Date: Sun, 5 Jul 2026 19:59:30 +0200
Subject: [PATCH 14/43] Refine BYOB host browser endpoint selection
Reduce Browser Settings host-browser choices to automatic and advertised debug endpoints, with a validated Custom endpoint input and faster inventory refresh.
Update Browser docs and regressions to document the custom endpoint workflow and CLI websocket selection.
---
docs/guides/a0-cli-connector.md | 8 +-
docs/guides/browser.md | 6 +-
plugins/_browser/AGENTS.md | 1 +
.../_browser/webui/browser-config-store.js | 100 +++++++++++++++++-
plugins/_browser/webui/config.html | 22 +++-
tests/test_browser_agent_regressions.py | 6 ++
6 files changed, 133 insertions(+), 10 deletions(-)
diff --git a/docs/guides/a0-cli-connector.md b/docs/guides/a0-cli-connector.md
index ee83105bc..f0812efc8 100644
--- a/docs/guides/a0-cli-connector.md
+++ b/docs/guides/a0-cli-connector.md
@@ -162,8 +162,9 @@ control starts when Agent Zero actually needs to use the browser.
> control.
The **Host browser** list in Browser settings comes from the connected local A0
-CLI, not from the Agent Zero Web UI server. If a newly authorized browser does
-not appear, restart or reconnect A0 CLI.
+CLI, not from the Agent Zero Web UI server. It shows Automatic, currently
+advertised debug endpoints, and **Custom endpoint**. If a newly authorized
+browser does not appear, restart or reconnect A0 CLI.
If the inspect checkbox is not enough for your browser build, launch it with an
explicit remote debugging port and a separate profile:
@@ -172,7 +173,8 @@ explicit remote debugging port and a separate profile:
opera --remote-debugging-port=9222 --user-data-dir="$HOME/.config/a0-opera-debug"
```
-Then pass the full DevTools websocket endpoint to A0 CLI:
+Then choose **Custom endpoint** in Browser settings, run `/browser ws://...` in
+A0 CLI, or pass the full DevTools websocket endpoint to A0 CLI:
```bash
export A0_HOST_BROWSER_REMOTE_DEBUGGING_ENDPOINTS="ws://127.0.0.1:9222/devtools/browser/..."
diff --git a/docs/guides/browser.md b/docs/guides/browser.md
index bc9255c1b..33d1b719f 100644
--- a/docs/guides/browser.md
+++ b/docs/guides/browser.md
@@ -156,7 +156,8 @@ Remote debugging pages:
| Chrome, Edge, Brave, Vivaldi, Chromium | `chrome://inspect/#remote-debugging` |
| Opera | `opera://inspect/#remote-debugging` |
-If a browser does not appear in the **Host browser** list after enabling remote
+The **Host browser** list shows Automatic, currently advertised debug endpoints,
+and **Custom endpoint**. If a browser does not appear after enabling remote
debugging, restart or reconnect the local A0 CLI. Restarting only the Agent Zero
Web UI server does not refresh the browser inventory; the list comes from the
connected CLI.
@@ -168,7 +169,8 @@ directory:
opera --remote-debugging-port=9222 --user-data-dir="$HOME/.config/a0-opera-debug"
```
-Then pass the full DevTools websocket endpoint to the CLI:
+Then choose **Custom endpoint** in Browser settings, or pass the full DevTools
+websocket endpoint to the CLI:
```bash
export A0_HOST_BROWSER_REMOTE_DEBUGGING_ENDPOINTS="ws://127.0.0.1:9222/devtools/browser/..."
diff --git a/plugins/_browser/AGENTS.md b/plugins/_browser/AGENTS.md
index c8f8a8c09..a0e570a9d 100644
--- a/plugins/_browser/AGENTS.md
+++ b/plugins/_browser/AGENTS.md
@@ -26,6 +26,7 @@
- Keep narrow WebUI Browser controls usable by grouping navigation with Annotate/settings above a full-width address bar.
- For Bring Your Own Browser with an existing host profile, `host_browser_selection` may target automatic CLI selection, a browser family/id, or an explicit CDP endpoint and must be forwarded to the connector runtime as `browser_selection`.
- Browser Settings must refresh connected A0 CLI host-browser inventory while the settings view is open so newly authorized endpoints appear without saving or reopening.
+- Browser Settings keeps the Host browser dropdown focused on automatic selection, advertised debug endpoints, and a validated Custom endpoint field instead of listing every installed local profile.
- Browser URL-intent handling must only claim web URL schemes and leave custom Agent Zero schemes to their owning surfaces.
- Prefer DOM/CDP browser actions with refs, selectors, frame-chain refs, and screenshots over viewport coordinate input. Coordinates remain a visual fallback.
- Do not hardcode user-specific browser paths or secrets.
diff --git a/plugins/_browser/webui/browser-config-store.js b/plugins/_browser/webui/browser-config-store.js
index 1f385430b..4f32fe0ac 100644
--- a/plugins/_browser/webui/browser-config-store.js
+++ b/plugins/_browser/webui/browser-config-store.js
@@ -10,7 +10,8 @@ const HOST_PROFILE_MODES = new Set(["existing", "agent"]);
const DEFAULT_MAX_OPEN_TABS = 32;
const MIN_MAX_OPEN_TABS = 1;
const HARD_MAX_OPEN_TABS = 50;
-const HOST_BROWSER_STATUS_REFRESH_MS = 3000;
+const HOST_BROWSER_STATUS_REFRESH_MS = 1000;
+const CUSTOM_HOST_BROWSER_SELECTION = "__custom_endpoint__";
function normalizePathList(value) {
const source = Array.isArray(value)
@@ -72,6 +73,39 @@ function normalizeHostBrowserSelection(value) {
return String(value || "").trim().toLowerCase().replace(/\s+/g, "_").slice(0, 200);
}
+function normalizeCustomHostBrowserEndpoint(value) {
+ const raw = String(value || "").trim();
+ if (!raw) return "";
+ const candidate = raw.includes("://") ? raw : `ws://${raw}`;
+ try {
+ const url = new URL(candidate);
+ if (!["ws:", "wss:"].includes(url.protocol) || !url.host || !url.pathname.startsWith("/devtools/browser/")) {
+ return "";
+ }
+ return normalizeHostBrowserSelection(`${url.protocol}//${url.host}${url.pathname}${url.search || ""}`);
+ } catch (_error) {
+ return "";
+ }
+}
+
+function isCustomHostBrowserEndpoint(value) {
+ return Boolean(normalizeCustomHostBrowserEndpoint(value));
+}
+
+function debugPortVersionUrl(value) {
+ const raw = String(value || "").trim();
+ if (!raw) return "";
+ const candidate = raw.includes("://") ? raw : `ws://${raw}`;
+ try {
+ const url = new URL(candidate);
+ if (!["ws:", "wss:"].includes(url.protocol) || !url.host) return "";
+ if (url.pathname && url.pathname !== "/") return "";
+ return `http://${url.host}/json/version`;
+ } catch (_error) {
+ return "";
+ }
+}
+
function normalizeBoolean(value, fallback = true) {
if (value === undefined || value === null || value === "") return fallback;
if (typeof value === "boolean") return value;
@@ -121,6 +155,8 @@ export const store = createStore("browserConfig", {
hostBrowserStatus: null,
hostBrowserStatusLoading: false,
hostBrowserStatusRefreshTimer: null,
+ hostBrowserCustomEndpoint: "",
+ hostBrowserCustomMode: false,
async init(config) {
this.bindConfig(config);
@@ -137,6 +173,8 @@ export const store = createStore("browserConfig", {
this.extensionDeleteLoadingPath = "";
this.hostBrowserStatus = null;
this.hostBrowserStatusLoading = false;
+ this.hostBrowserCustomEndpoint = "";
+ this.hostBrowserCustomMode = false;
},
startHostBrowserStatusRefresh() {
@@ -158,6 +196,9 @@ export const store = createStore("browserConfig", {
if (!safeConfig) return;
if (this.config === safeConfig) return;
this.config = safeConfig;
+ if (isCustomHostBrowserEndpoint(safeConfig.host_browser_selection)) {
+ this.hostBrowserCustomEndpoint = safeConfig.host_browser_selection;
+ }
},
setAutofocusActivePage(enabled) {
@@ -215,29 +256,82 @@ export const store = createStore("browserConfig", {
? connector.available_browsers
: [];
for (const browser of advertised) {
- const value = normalizeHostBrowserSelection(browser?.id || browser?.family || browser?.cdp_endpoint);
+ const value = normalizeCustomHostBrowserEndpoint(browser?.cdp_endpoint || browser?.id);
if (!value || seen.has(value)) continue;
seen.add(value);
const label = browser?.label || hostBrowserFamilyLabel(browser?.family || value);
const status = browser?.status ? ` - ${hostBrowserStatusLabel(browser.status)}` : "";
options.push({ value, label: `${label}${status}` });
}
- const fallbackValue = normalizeHostBrowserSelection(connector?.browser_id || connector?.browser_family);
+ const fallbackValue = normalizeCustomHostBrowserEndpoint(connector?.cdp_endpoint || connector?.browser_id);
if (fallbackValue && !seen.has(fallbackValue)) {
seen.add(fallbackValue);
const label = connector?.browser_label || hostBrowserFamilyLabel(connector?.browser_family || fallbackValue);
options.push({ value: fallbackValue, label });
}
}
+ const selected = normalizeHostBrowserSelection(this.config?.host_browser_selection);
+ if (selected && !seen.has(selected) && !isCustomHostBrowserEndpoint(selected)) {
+ seen.add(selected);
+ options.push({ value: selected, label: `Saved: ${selected}` });
+ }
+ options.push({ value: CUSTOM_HOST_BROWSER_SELECTION, label: "Custom endpoint" });
return options;
},
+ hostBrowserSelectValue() {
+ if (this.hostBrowserCustomMode) return CUSTOM_HOST_BROWSER_SELECTION;
+ const selected = normalizeHostBrowserSelection(this.config?.host_browser_selection);
+ if (!selected) return "";
+ if (this.hostBrowserOptions().some((option) => option.value === selected)) return selected;
+ if (isCustomHostBrowserEndpoint(selected)) return CUSTOM_HOST_BROWSER_SELECTION;
+ return selected;
+ },
+
setHostBrowserSelection(value) {
const safeConfig = ensureConfig(this.config);
if (!safeConfig) return;
+ if (value === CUSTOM_HOST_BROWSER_SELECTION) {
+ this.hostBrowserCustomMode = true;
+ if (isCustomHostBrowserEndpoint(safeConfig.host_browser_selection)) {
+ this.hostBrowserCustomEndpoint = safeConfig.host_browser_selection;
+ } else {
+ safeConfig.host_browser_selection = "";
+ }
+ return;
+ }
+ this.hostBrowserCustomMode = false;
safeConfig.host_browser_selection = normalizeHostBrowserSelection(value);
},
+ showCustomHostBrowserEndpoint() {
+ return this.hostBrowserSelectValue() === CUSTOM_HOST_BROWSER_SELECTION;
+ },
+
+ setCustomHostBrowserEndpoint(value) {
+ this.hostBrowserCustomMode = true;
+ this.hostBrowserCustomEndpoint = String(value || "").trim();
+ const safeConfig = ensureConfig(this.config);
+ if (!safeConfig) return;
+ const endpoint = normalizeCustomHostBrowserEndpoint(this.hostBrowserCustomEndpoint);
+ if (endpoint || !this.hostBrowserCustomEndpoint) {
+ safeConfig.host_browser_selection = endpoint;
+ }
+ },
+
+ customHostBrowserEndpointDiagnostic() {
+ if (!this.hostBrowserCustomEndpoint) {
+ return "Paste a ws://.../devtools/browser/... endpoint from the browser inspect page.";
+ }
+ const endpoint = normalizeCustomHostBrowserEndpoint(this.hostBrowserCustomEndpoint);
+ if (endpoint) return `Using ${endpoint}`;
+ const versionUrl = debugPortVersionUrl(this.hostBrowserCustomEndpoint);
+ if (versionUrl) {
+ return `This looks like a debug port. Open ${versionUrl} and copy webSocketDebuggerUrl.`;
+ }
+ return "Endpoint must be a ws:// or wss:// URL ending in /devtools/browser/...";
+ },
+
hostBrowserProfileModeLabel() {
const value = this.config?.host_browser_profile_mode || "existing";
if (value === "agent") return "Clean Agent Profile";
diff --git a/plugins/_browser/webui/config.html b/plugins/_browser/webui/config.html
index 79851960c..40cba5d37 100644
--- a/plugins/_browser/webui/config.html
+++ b/plugins/_browser/webui/config.html
@@ -72,14 +72,32 @@
>
Date: Sun, 5 Jul 2026 20:11:19 +0200
Subject: [PATCH 15/43] Normalize MCP stdio command lines
Accept shell-style local MCP command values from the manager before spawning stdio clients, so configs like uvx workspace-mcp resolve to the correct executable and args.
Split collapsed option/value argument lines in the MCP manager, document the input contract, and add a regression for the google_workspace MCP shape.
---
helpers/mcp_handler.py | 52 +++++++++++++++++++
helpers/mcp_handler.py.dox.md | 6 ++-
tests/test_mcp_handler_multimodal.py | 25 +++++++++
webui/components/settings/AGENTS.md | 1 +
.../settings/mcp/client/mcp-servers-store.js | 2 +-
5 files changed, 84 insertions(+), 2 deletions(-)
diff --git a/helpers/mcp_handler.py b/helpers/mcp_handler.py
index b1f888ec1..e30c3916c 100644
--- a/helpers/mcp_handler.py
+++ b/helpers/mcp_handler.py
@@ -21,6 +21,7 @@ from contextlib import AsyncExitStack
from shutil import which
from datetime import timedelta
import json
+import shlex
import uuid
from helpers import errors
from helpers import settings
@@ -112,6 +113,44 @@ def _normalize_disabled_tools(value: Any) -> list[str]:
return [str(item).strip() for item in value if str(item).strip()]
+def _split_stdio_command(command: Any) -> tuple[str, list[str]]:
+ text = str(command or "").strip()
+ if not text:
+ return "", []
+ try:
+ parts = shlex.split(text)
+ except ValueError:
+ return text, []
+ if not parts:
+ return "", []
+ return parts[0], parts[1:]
+
+
+def _split_stdio_arg_fragment(arg: str) -> list[str]:
+ try:
+ parts = shlex.split(arg)
+ except ValueError:
+ return [arg]
+ if len(parts) <= 1:
+ return parts or []
+ if parts[0].startswith("-") and "=" not in parts[0]:
+ return parts
+ if any(part.startswith("-") for part in parts[1:]):
+ return parts
+ return [arg]
+
+
+def _normalize_stdio_args(value: Any) -> list[str]:
+ if not isinstance(value, list):
+ return []
+ args: list[str] = []
+ for item in value:
+ text = str(item).strip()
+ if text:
+ args.extend(_split_stdio_arg_fragment(text))
+ return args
+
+
def initialize_mcp(mcp_servers_config: str):
if not MCPConfig.get_instance().is_initialized():
try:
@@ -649,6 +688,15 @@ class MCPServerLocal(BaseModel):
def update(self, config: dict[str, Any]) -> "MCPServerLocal":
with self.__lock:
+ command = self.command
+ command_args: list[str] = []
+ args = list(self.args)
+ if "command" in config:
+ command, command_args = _split_stdio_command(config.get("command"))
+ args = [*command_args, *args]
+ if "args" in config:
+ args = [*command_args, *_normalize_stdio_args(config.get("args"))]
+
for key, value in config.items():
if key in [
"name",
@@ -665,11 +713,15 @@ class MCPServerLocal(BaseModel):
"disabled_tools",
"scope",
]:
+ if key in ["command", "args"]:
+ continue
if key == "name":
value = normalize_name(value)
if key == "disabled_tools":
value = _normalize_disabled_tools(value)
setattr(self, key, value)
+ self.command = command
+ self.args = args
return self
async def initialize(self) -> "MCPServerLocal":
diff --git a/helpers/mcp_handler.py.dox.md b/helpers/mcp_handler.py.dox.md
index 4faff9fe1..1cab10612 100644
--- a/helpers/mcp_handler.py.dox.md
+++ b/helpers/mcp_handler.py.dox.md
@@ -66,6 +66,9 @@
- `_is_streaming_http_type(server_type: str) -> bool`: Check if the server type is a streaming HTTP variant.
- `_split_qualified_tool_name(tool_name: str) -> tuple[str, str]`: Split `server.tool` names while preserving dots inside MCP tool names.
- `_normalize_disabled_tools(value: Any) -> list[str]`: Normalize the optional per-server disabled tool list.
+- `_split_stdio_command(command: Any) -> tuple[str, list[str]]`: Split shell-style local MCP command lines into an executable plus leading arguments.
+- `_split_stdio_arg_fragment(arg: str) -> list[str]`: Split collapsed option/value argument fragments while preserving obvious single values with spaces.
+- `_normalize_stdio_args(value: Any) -> list[str]`: Normalize local MCP argument lists after manager/raw JSON parsing.
- `initialize_mcp(mcp_servers_config: str)`
- Notable constants/configuration names: `DEFAULT_MCP_SERVERS_CONFIG`, `MCP_MEDIA_TOKENS_ESTIMATE`, `MAX_MCP_RESOURCE_TEXT_CHARS`, `MCP_SESSION_CLEANUP_TIMEOUT_SECONDS`, `MCP_OPERATION_TIMEOUT_GRACE_SECONDS`, `T`.
@@ -81,12 +84,13 @@
- MCP tool names are qualified as `server_name.tool_name`; server names are normalized without dots, and the tool portion may contain dots.
- Servers may define `disabled_tools` as a list of MCP tool names. Disabled tools are omitted from agent-facing prompts, status counts, `has_tool`, and calls, while detail views can still retrieve them through `get_all_tools()` with a `disabled` flag so users can re-enable them.
- Server-specific `init_timeout` and `tool_timeout` override global MCP client timeout settings for list-tools and call-tool operations.
+- Local stdio server configs accept either strict MCP JSON (`command: "uvx", args: [...]`) or manager-style command lines (`command: "uvx package"`) and normalize them before spawning the process.
- MCP image and image-resource content is materialized to scoped artifact files and returned both as model-visible image attachments and as path metadata (`attachments`/`media_paths`) for downstream delivery.
- MCP config locks must not be held across awaited server initialization or tool-call operations. Slow or wedged MCP servers must not block status reads, prompt construction, unrelated MCP servers, or later tool calls through the shared config lock.
- MCP client session work runs inside disposable isolated `DeferredTask` workers with an outer timeout. Normal `AsyncExitStack` cleanup is also bounded; if cleanup or transport shutdown does not finish, the operation reports failure or warning while Agent Zero keeps control of the agent loop.
- Server status marks initialized server objects with cached initialization errors as disconnected, even if the config object exists.
- Observed side-effect areas: filesystem writes, network calls, WebSocket state, settings/state persistence, secret handling.
-- Imported dependency areas include: `abc`, `anyio.streams.memory`, `asyncio`, `contextlib`, `datetime`, `helpers`, `helpers.defer`, `helpers.log`, `helpers.print_style`, `helpers.tool`, `httpx`, `json`, `mcp`, `mcp.client.sse`, `mcp.client.stdio`, `mcp.client.streamable_http`, `mcp.shared.message`.
+- Imported dependency areas include: `abc`, `anyio.streams.memory`, `asyncio`, `contextlib`, `datetime`, `helpers`, `helpers.defer`, `helpers.log`, `helpers.print_style`, `helpers.tool`, `httpx`, `json`, `mcp`, `mcp.client.sse`, `mcp.client.stdio`, `mcp.client.streamable_http`, `mcp.shared.message`, `shlex`.
## Key Concepts
diff --git a/tests/test_mcp_handler_multimodal.py b/tests/test_mcp_handler_multimodal.py
index 19b4ca123..fd402dcc7 100644
--- a/tests/test_mcp_handler_multimodal.py
+++ b/tests/test_mcp_handler_multimodal.py
@@ -345,6 +345,31 @@ def test_mcp_disabled_tools_are_hidden_from_agent_paths_but_visible_in_detail(mc
assert malformed_config.servers[0].disabled_tools == []
+def test_mcp_local_server_accepts_manager_style_command_lines(mcp_handler_module):
+ module, _tmp_path = mcp_handler_module
+
+ server = module.MCPServerLocal(
+ {
+ "name": "google_workspace",
+ "command": "uvx workspace-mcp",
+ "args": [
+ "--tool-tier core",
+ "/tmp/path with spaces",
+ "--label=Two Words",
+ ],
+ }
+ )
+
+ assert server.command == "uvx"
+ assert server.args == [
+ "workspace-mcp",
+ "--tool-tier",
+ "core",
+ "/tmp/path with spaces",
+ "--label=Two Words",
+ ]
+
+
def test_mcp_client_call_tool_uses_server_tool_timeout(mcp_handler_module, monkeypatch):
module, _tmp_path = mcp_handler_module
session_timeouts = []
diff --git a/webui/components/settings/AGENTS.md b/webui/components/settings/AGENTS.md
index 2f5529227..463462a05 100644
--- a/webui/components/settings/AGENTS.md
+++ b/webui/components/settings/AGENTS.md
@@ -18,6 +18,7 @@
- Preserve Store Gating and modal footer conventions in settings components.
- MCP manager tool toggles write `disabled_tools` into the draft JSON and require Apply before changing the running MCP tool set.
- Confirmed MCP server removals apply immediately and refresh server status; other MCP manager draft edits still require Apply.
+- MCP manager local command forms accept shell-style command and argument lines; quote argument values that intentionally contain spaces.
## Work Guidance
diff --git a/webui/components/settings/mcp/client/mcp-servers-store.js b/webui/components/settings/mcp/client/mcp-servers-store.js
index 0963438c0..3d73d60d8 100644
--- a/webui/components/settings/mcp/client/mcp-servers-store.js
+++ b/webui/components/settings/mcp/client/mcp-servers-store.js
@@ -117,7 +117,7 @@ function parseArgsText(text) {
const parsed = JSON.parse(raw);
return Array.isArray(parsed) ? parsed.map((item) => String(item)) : [];
}
- return raw.split(/\r?\n/).map((line) => line.trim()).filter(Boolean);
+ return raw.split(/\r?\n/).flatMap((line) => splitCommandLine(line.trim())).filter(Boolean);
}
function formatArgsText(value) {
From 8871bf5001c7f4e6f41d2cf4d5092bfcb1018c4c Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Mon, 6 Jul 2026 15:28:57 +0200
Subject: [PATCH 16/43] Update launcher README links to v1.2
Point the root README's A0 Launcher downloads and release note link at the published v1.2 assets without moving the Agent Zero v2.2 tag.
---
README.md | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/README.md b/README.md
index 9a88f34f0..ac127322c 100644
--- a/README.md
+++ b/README.md
@@ -49,10 +49,10 @@ The desktop **A0 Launcher** is the recommended way to install Agent Zero on a pe
| Architecture | macOS | Linux | Windows |
| --- | --- | --- | --- |
-| x86 | [Mac Intel](https://github.com/agent0ai/a0-launcher/releases/download/v1.1/a0-launcher-1.1-macos-x64.dmg) | [Linux x86](https://github.com/agent0ai/a0-launcher/releases/download/v1.1/a0-launcher-1.1-linux-x64.AppImage) | [Windows x86](https://github.com/agent0ai/a0-launcher/releases/download/v1.1/a0-launcher-1.1-windows-x64.exe) |
-| ARM64 | [Mac Apple Silicon](https://github.com/agent0ai/a0-launcher/releases/download/v1.1/a0-launcher-1.1-macos-arm64.dmg) | [Linux ARM64](https://github.com/agent0ai/a0-launcher/releases/download/v1.1/a0-launcher-1.1-linux-arm64.AppImage) | [Windows ARM64](https://github.com/agent0ai/a0-launcher/releases/download/v1.1/a0-launcher-1.1-windows-arm64.exe) |
+| x86 | [Mac Intel](https://github.com/agent0ai/a0-launcher/releases/download/v1.2/a0-launcher-1.2-macos-x64.dmg) | [Linux x86](https://github.com/agent0ai/a0-launcher/releases/download/v1.2/a0-launcher-1.2-linux-x64.AppImage) | [Windows x86](https://github.com/agent0ai/a0-launcher/releases/download/v1.2/a0-launcher-1.2-windows-x64.exe) |
+| ARM64 | [Mac Apple Silicon](https://github.com/agent0ai/a0-launcher/releases/download/v1.2/a0-launcher-1.2-macos-arm64.dmg) | [Linux ARM64](https://github.com/agent0ai/a0-launcher/releases/download/v1.2/a0-launcher-1.2-linux-arm64.AppImage) | [Windows ARM64](https://github.com/agent0ai/a0-launcher/releases/download/v1.2/a0-launcher-1.2-windows-arm64.exe) |
-See the [A0 Launcher v1.1 release](https://github.com/agent0ai/a0-launcher/releases/tag/v1.1) for release notes and updater metadata. See the [Launcher guide](./docs/guides/launcher.md) for the first-run walkthrough.
+See the [A0 Launcher v1.2 release](https://github.com/agent0ai/a0-launcher/releases/tag/v1.2) for release notes and updater metadata. See the [Launcher guide](./docs/guides/launcher.md) for the first-run walkthrough.
## A0 Install
From 101d84551de2f1fef3f69572b986ca96fb292a43 Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Mon, 6 Jul 2026 15:28:57 +0200
Subject: [PATCH 17/43] Update launcher README links to v1.2
Point the root README's A0 Launcher downloads and release note link at the published v1.2 assets without moving the Agent Zero v2.2 tag.
---
README.md | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/README.md b/README.md
index c5eb3f50c..d3d5ebfc8 100644
--- a/README.md
+++ b/README.md
@@ -49,10 +49,10 @@ The desktop **A0 Launcher** is the recommended way to install Agent Zero on a pe
| Architecture | macOS | Linux | Windows |
| --- | --- | --- | --- |
-| x86 | [Mac Intel](https://github.com/agent0ai/a0-launcher/releases/download/v1.1/a0-launcher-1.1-macos-x64.dmg) | [Linux x86](https://github.com/agent0ai/a0-launcher/releases/download/v1.1/a0-launcher-1.1-linux-x64.AppImage) | [Windows x86](https://github.com/agent0ai/a0-launcher/releases/download/v1.1/a0-launcher-1.1-windows-x64.exe) |
-| ARM64 | [Mac Apple Silicon](https://github.com/agent0ai/a0-launcher/releases/download/v1.1/a0-launcher-1.1-macos-arm64.dmg) | [Linux ARM64](https://github.com/agent0ai/a0-launcher/releases/download/v1.1/a0-launcher-1.1-linux-arm64.AppImage) | [Windows ARM64](https://github.com/agent0ai/a0-launcher/releases/download/v1.1/a0-launcher-1.1-windows-arm64.exe) |
+| x86 | [Mac Intel](https://github.com/agent0ai/a0-launcher/releases/download/v1.2/a0-launcher-1.2-macos-x64.dmg) | [Linux x86](https://github.com/agent0ai/a0-launcher/releases/download/v1.2/a0-launcher-1.2-linux-x64.AppImage) | [Windows x86](https://github.com/agent0ai/a0-launcher/releases/download/v1.2/a0-launcher-1.2-windows-x64.exe) |
+| ARM64 | [Mac Apple Silicon](https://github.com/agent0ai/a0-launcher/releases/download/v1.2/a0-launcher-1.2-macos-arm64.dmg) | [Linux ARM64](https://github.com/agent0ai/a0-launcher/releases/download/v1.2/a0-launcher-1.2-linux-arm64.AppImage) | [Windows ARM64](https://github.com/agent0ai/a0-launcher/releases/download/v1.2/a0-launcher-1.2-windows-arm64.exe) |
-See the [A0 Launcher v1.1 release](https://github.com/agent0ai/a0-launcher/releases/tag/v1.1) for release notes and updater metadata. See the [Launcher guide](./docs/guides/launcher.md) for the first-run walkthrough.
+See the [A0 Launcher v1.2 release](https://github.com/agent0ai/a0-launcher/releases/tag/v1.2) for release notes and updater metadata. See the [Launcher guide](./docs/guides/launcher.md) for the first-run walkthrough.
## A0 Install
From 9bb3cb9469c7b9fbc6a1243f4b991c68deb0d91c Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Tue, 7 Jul 2026 13:56:26 +0200
Subject: [PATCH 18/43] Render parallel code execution as EXE
Use the code_exe log type for code_execution_tool children launched through parallel so the WebUI reuses the existing code execution renderer.
Add a focused regression test and document the parallel child-log contract.
---
helpers/parallel_tools.py | 13 +++++++++
helpers/parallel_tools.py.dox.md | 1 +
tests/test_parallel_tool.py | 46 ++++++++++++++++++++++++++++++++
3 files changed, 60 insertions(+)
diff --git a/helpers/parallel_tools.py b/helpers/parallel_tools.py
index e9d830bef..5b62bdc5b 100644
--- a/helpers/parallel_tools.py
+++ b/helpers/parallel_tools.py
@@ -600,6 +600,19 @@ def _log_parallel_child_started(agent: "Agent", job: ParallelJob) -> None:
)
return
+ if job.tool_name == "code_execution_tool":
+ runtime = job.tool_args.get("runtime", "unknown")
+ session = job.tool_args.get("session", None)
+ session_text = f"[{session}] " if session or session == 0 else ""
+ job.log_item = agent.context.log.log(
+ type="code_exe",
+ heading=f"icon://terminal {session_text}code_execution_tool - {runtime}",
+ content="",
+ kvps=job.tool_args,
+ id=job.log_id,
+ )
+ return
+
heading = f"icon://construction {agent.agent_name}: Using tool '{job.tool_name}'"
job.log_item = agent.context.log.log(
type="tool",
diff --git a/helpers/parallel_tools.py.dox.md b/helpers/parallel_tools.py.dox.md
index 39e65c4cc..2aedf01ab 100644
--- a/helpers/parallel_tools.py.dox.md
+++ b/helpers/parallel_tools.py.dox.md
@@ -31,6 +31,7 @@
- Direct tool background context cleanup removes both the in-memory context and any transient chat folder left on disk.
- Parent-visible child log items are created for each wrapped call so the WebUI can inspect concurrent children separately while the wrapper result remains model-history-only.
- Child tool logs mirror normal tool-call visible args; job ids remain available through wrapper results and prompt extras rather than visible process-step args.
+- Wrapped `code_execution_tool` child logs use the normal `code_exe` WebUI message type and terminal heading so they render like direct code execution calls.
- Job IDs are stable handles for later await, collect, or cancel operations.
- Prompt extras must stay bounded and expose only job IDs, tool names, status, and compact result/error summaries.
diff --git a/tests/test_parallel_tool.py b/tests/test_parallel_tool.py
index 587604aa8..bfb65e66e 100644
--- a/tests/test_parallel_tool.py
+++ b/tests/test_parallel_tool.py
@@ -427,6 +427,52 @@ async def test_parallel_direct_tool_jobs_log_normal_tool_metadata(monkeypatch) -
assert agent.context.log.items[0].kvps == {"seconds": 1, "_tool_name": "wait"}
+@pytest.mark.asyncio
+async def test_parallel_code_execution_child_uses_code_exe_log_type(monkeypatch) -> None:
+ class FakeDeferredTask:
+ def __init__(self, thread_name=None) -> None:
+ self.thread_name = thread_name
+
+ def start_task(self, func, *args):
+ return self
+
+ def is_ready(self):
+ return False
+
+ def is_alive(self):
+ return True
+
+ def kill(self):
+ pass
+
+ monkeypatch.setattr(parallel_tools, "DeferredTask", FakeDeferredTask)
+ agent = _FakeAgent()
+
+ jobs = await parallel_tools.start_parallel_jobs(
+ agent, # type: ignore[arg-type]
+ [
+ parallel_tools.NormalizedToolCall(
+ index=0,
+ tool_name="code_execution_tool",
+ tool_args={
+ "runtime": "terminal",
+ "session": 0,
+ "code": "pwd",
+ },
+ )
+ ],
+ )
+
+ assert jobs[0].kind == "tool"
+ assert agent.context.log.items[0].type == "code_exe"
+ assert agent.context.log.items[0].heading == "icon://terminal [0] code_execution_tool - terminal"
+ assert agent.context.log.items[0].kvps == {
+ "runtime": "terminal",
+ "session": 0,
+ "code": "pwd",
+ }
+
+
@pytest.mark.asyncio
async def test_parallel_tool_keeps_wrapper_out_of_visible_log() -> None:
from tools.parallel import ParallelTool
From 96fd5fc2b22f40f2267fe55a04ae2b35f35a254b Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Tue, 7 Jul 2026 15:15:36 +0200
Subject: [PATCH 19/43] Ignore browser key handling for local editable targets
Prevent Browser keyboard capture from preventing chat composer typing when focus is in editable fields (including contenteditable).
---
plugins/_browser/webui/browser-store.js | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/plugins/_browser/webui/browser-store.js b/plugins/_browser/webui/browser-store.js
index e043751d6..67c151383 100644
--- a/plugins/_browser/webui/browser-store.js
+++ b/plugins/_browser/webui/browser-store.js
@@ -1986,6 +1986,7 @@ const model = {
},
handleKeydown(event) {
+ if (isLocalEditableTarget(event?.target)) return;
const annotateShortcut = event?.key === "." && (event.metaKey || event.ctrlKey) && !event.altKey;
if (annotateShortcut && this._surfaceMounted) {
event.preventDefault();
@@ -2700,8 +2701,7 @@ const model = {
const contextId = this.normalizeContextId(this.activeBrowserContextId || this.contextId);
if (!contextId || !this.activeBrowserId) return;
if (event.ctrlKey || event.metaKey || event.altKey) return;
- const editable = ["INPUT", "TEXTAREA", "SELECT"].includes(event.target?.tagName);
- if (editable) return;
+ if (isLocalEditableTarget(event?.target)) return;
event.preventDefault();
const printable = event.key && event.key.length === 1;
await websocket.emit("browser_viewer_input", {
From 67895f7b44033397c8e151279e9425f4a5714f34 Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Tue, 7 Jul 2026 16:52:27 +0200
Subject: [PATCH 20/43] Default direct Venice to chat completions
Set the shipped direct Venice chat provider default to use Agent Zero's chat-completions transport while preserving the existing Venice API base and disabled Venice system prompt injection.\n\nAdd a focused regression check that direct venice gets the chat default, a0_venice and embeddings stay untouched, and explicit model kwargs can still override the provider default.
---
conf/model_providers.yaml | 1 +
tests/test_model_config_api_keys.py | 28 ++++++++++++++++++++++++++++
2 files changed, 29 insertions(+)
diff --git a/conf/model_providers.yaml b/conf/model_providers.yaml
index 541a4888b..ba3f0ea48 100644
--- a/conf/model_providers.yaml
+++ b/conf/model_providers.yaml
@@ -173,6 +173,7 @@ chat:
models_list:
endpoint_url: "https://api.venice.ai/api/v1/models"
kwargs:
+ a0_api_mode: chat
api_base: https://api.venice.ai/api/v1
venice_parameters:
include_venice_system_prompt: false
diff --git a/tests/test_model_config_api_keys.py b/tests/test_model_config_api_keys.py
index 18b41f4ff..e1f62d2f0 100644
--- a/tests/test_model_config_api_keys.py
+++ b/tests/test_model_config_api_keys.py
@@ -270,6 +270,34 @@ def test_ollama_cloud_provider_config_requires_key_and_base_url():
assert "api_key_mode" not in ollama_cloud
+def test_direct_venice_chat_provider_defaults_to_chat_completions(monkeypatch):
+ import yaml
+
+ monkeypatch.setattr(models, "get_api_key", lambda provider: "None")
+
+ provider_path = PROJECT_ROOT / "conf/model_providers.yaml"
+ provider_config = yaml.safe_load(provider_path.read_text(encoding="utf-8"))
+
+ venice = provider_config["chat"]["venice"]
+ assert venice["kwargs"]["a0_api_mode"] == "chat"
+ assert venice["kwargs"]["api_base"] == "https://api.venice.ai/api/v1"
+ assert venice["kwargs"]["venice_parameters"] == {
+ "include_venice_system_prompt": False
+ }
+ assert "a0_api_mode" not in provider_config["chat"]["a0_venice"]["kwargs"]
+ assert "a0_api_mode" not in provider_config["embedding"]["venice"]["kwargs"]
+
+ model = models.get_chat_model("venice", "llama-3.3-70b")
+ assert model.kwargs["a0_api_mode"] == "chat"
+
+ custom = models.get_chat_model(
+ "venice",
+ "llama-3.3-70b",
+ a0_api_mode="responses",
+ )
+ assert custom.kwargs["a0_api_mode"] == "responses"
+
+
def test_missing_api_key_banner_does_not_include_auto_modal_metadata(monkeypatch):
from plugins._model_config.helpers import model_config
From ad148f24cc1bcf66c8f8ec9caece6f0909f9a39d Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Tue, 7 Jul 2026 16:59:49 +0200
Subject: [PATCH 21/43] Fix Responses state after history compaction
Clear the active Responses provider continuation when automatic history compression or manual chat compaction rewrites local history, while preserving stored response IDs for cleanup.
Add focused regressions for both compression paths so compacted chats do not keep stale provider-side context.
---
extensions/python/message_loop_end/AGENTS.md | 1 +
.../message_loop_end/_10_organize_history.py | 10 ++-
.../_90_organize_history_wait.py | 7 +-
helpers/history.py | 22 ++++++
helpers/history.py.dox.md | 2 +
plugins/_chat_compaction/AGENTS.md | 1 +
plugins/_chat_compaction/helpers/compactor.py | 3 +-
tests/test_chat_compaction.py | 79 +++++++++++++++++++
tests/test_history_compression_wait.py | 34 ++++++++
9 files changed, 155 insertions(+), 4 deletions(-)
diff --git a/extensions/python/message_loop_end/AGENTS.md b/extensions/python/message_loop_end/AGENTS.md
index 8927890a9..2200c4aa3 100644
--- a/extensions/python/message_loop_end/AGENTS.md
+++ b/extensions/python/message_loop_end/AGENTS.md
@@ -12,6 +12,7 @@
- Preserve history consistency before saving chats.
- Do not skip persistence for successful loops unless the hook contract explicitly permits it.
+- History compression that rewrites local history must clear the active Responses provider continuation while preserving stored response IDs for cleanup.
## Work Guidance
diff --git a/extensions/python/message_loop_end/_10_organize_history.py b/extensions/python/message_loop_end/_10_organize_history.py
index d91c9ef80..0cc205bd6 100644
--- a/extensions/python/message_loop_end/_10_organize_history.py
+++ b/extensions/python/message_loop_end/_10_organize_history.py
@@ -1,10 +1,18 @@
from helpers.extension import Extension
from agent import LoopData
from helpers.defer import DeferredTask, THREAD_BACKGROUND
+from helpers.history import clear_responses_provider_state
DATA_NAME_TASK = "_organize_history_task"
+async def compress_history(agent) -> bool:
+ compressed = bool(await agent.history.compress())
+ if compressed:
+ clear_responses_provider_state(agent)
+ return compressed
+
+
class OrganizeHistory(Extension):
async def execute(self, loop_data: LoopData = LoopData(), **kwargs):
if not self.agent:
@@ -17,6 +25,6 @@ class OrganizeHistory(Extension):
# start task
task = DeferredTask(thread_name=THREAD_BACKGROUND)
- task.start_task(self.agent.history.compress)
+ task.start_task(compress_history, self.agent)
# set to agent to be able to wait for it
self.agent.set_data(DATA_NAME_TASK, task)
diff --git a/extensions/python/message_loop_prompts_before/_90_organize_history_wait.py b/extensions/python/message_loop_prompts_before/_90_organize_history_wait.py
index b2c925fe8..08966d2b1 100644
--- a/extensions/python/message_loop_prompts_before/_90_organize_history_wait.py
+++ b/extensions/python/message_loop_prompts_before/_90_organize_history_wait.py
@@ -1,6 +1,9 @@
from helpers.extension import Extension
from agent import LoopData
-from extensions.python.message_loop_end._10_organize_history import DATA_NAME_TASK
+from extensions.python.message_loop_end._10_organize_history import (
+ DATA_NAME_TASK,
+ compress_history,
+)
from helpers.defer import DeferredTask, THREAD_BACKGROUND
MAX_SYNC_COMPRESSION_PASSES = 64
@@ -33,7 +36,7 @@ class OrganizeHistoryWait(Extension):
else:
# no task was running, start and wait
self.agent.context.log.set_progress("Compressing history...")
- compressed = await self.agent.history.compress()
+ compressed = await compress_history(self.agent)
after_tokens = self.agent.history.get_tokens()
if not compressed or after_tokens >= before_tokens:
diff --git a/helpers/history.py b/helpers/history.py
index c579c501d..1ff802d87 100644
--- a/helpers/history.py
+++ b/helpers/history.py
@@ -723,6 +723,28 @@ def output_text(messages: list[OutputMessage], ai_label="ai", human_label="human
return "\n".join(_stringify_output(o, ai_label, human_label) for o in messages)
+def clear_responses_provider_state(agent) -> None:
+ key = getattr(agent, "DATA_NAME_RESPONSES_STATE", "responses_state")
+ get_data = getattr(agent, "get_data", None)
+ set_data = getattr(agent, "set_data", None)
+ if not callable(get_data) or not callable(set_data):
+ return
+
+ state = get_data(key)
+ if not isinstance(state, dict):
+ return
+
+ state = dict(state)
+ removed = False
+ for field in ("response_id", "previous_response_id"):
+ if field in state:
+ state.pop(field, None)
+ removed = True
+
+ if removed:
+ set_data(key, state)
+
+
def _merge_outputs(a: MessageContent, b: MessageContent) -> MessageContent:
if isinstance(a, str) and isinstance(b, str):
return a + "\n" + b
diff --git a/helpers/history.py.dox.md b/helpers/history.py.dox.md
index 0e07d4a9e..53a7149e0 100644
--- a/helpers/history.py.dox.md
+++ b/helpers/history.py.dox.md
@@ -65,6 +65,7 @@
- `group_messages_abab(messages: list[BaseMessage]) -> list[BaseMessage]`
- `output_langchain(messages: list[OutputMessage])`
- `output_text(messages: list[OutputMessage], ai_label=..., human_label=...)`
+- `clear_responses_provider_state(agent) -> None`
- `_merge_outputs(a: MessageContent, b: MessageContent) -> MessageContent`
- `_merge_properties(a: Dict[str, MessageContent], b: Dict[str, MessageContent]) -> Dict[str, MessageContent]`
- `_is_raw_message(obj: object) -> bool`
@@ -77,6 +78,7 @@
- Helper modules own reusable framework APIs and must preserve public callers unless all callers, tests, and docs are updated together.
- Update this file whenever public functions, classes, persistence behavior, path/security assumptions, side effects, or cross-module contracts change.
+- `clear_responses_provider_state(agent)` removes the active provider continuation IDs after local history rewrites while preserving stored response ID lists for later cleanup.
- Observed side-effect areas: filesystem writes, filesystem deletion, model calls, plugin state, settings/state persistence, secret handling.
- Imported dependency areas include: `abc`, `asyncio`, `collections`, `collections.abc`, `enum`, `helpers`, `json`, `langchain_core.messages`, `math`, `plugins._model_config.helpers.model_config`, `typing`, `uuid`.
diff --git a/plugins/_chat_compaction/AGENTS.md b/plugins/_chat_compaction/AGENTS.md
index ba56b0033..7f318836c 100644
--- a/plugins/_chat_compaction/AGENTS.md
+++ b/plugins/_chat_compaction/AGENTS.md
@@ -19,6 +19,7 @@
- Backup JSON and transcript artifacts must remain UTF-8 writable when chat content contains malformed Unicode such as lone surrogates.
- Keep generated summaries bounded by configured model and token limits.
- Preserve loaded skill names from `skill_instructions` metadata without copying full skill bodies into compacted summaries.
+- After replacing local history, clear the active Responses provider continuation while preserving stored response IDs for cleanup.
- Do not discard original context data unless the compaction flow explicitly owns that behavior.
## Work Guidance
diff --git a/plugins/_chat_compaction/helpers/compactor.py b/plugins/_chat_compaction/helpers/compactor.py
index b61cd7eed..f0a56c225 100644
--- a/plugins/_chat_compaction/helpers/compactor.py
+++ b/plugins/_chat_compaction/helpers/compactor.py
@@ -5,7 +5,7 @@ from collections import deque
import models as models_module
from agent import Agent
from helpers import files, tokens
-from helpers.history import History, output_text
+from helpers.history import History, clear_responses_provider_state, output_text
from helpers.persist_chat import (
export_json_chat,
get_chat_folder_path,
@@ -145,6 +145,7 @@ async def run_compaction(
agent.history = History(agent=agent)
agent.history.add_message(ai=True, content=compacted_content)
+ clear_responses_provider_state(agent)
# Clear subordinate chain
agent.data.pop(Agent.DATA_NAME_SUBORDINATE, None)
diff --git a/tests/test_chat_compaction.py b/tests/test_chat_compaction.py
index fa8358c57..f9678169f 100644
--- a/tests/test_chat_compaction.py
+++ b/tests/test_chat_compaction.py
@@ -43,6 +43,49 @@ class _RecordingModel:
return f"summary-{len(self.user_messages)}", None
+class _CompactionHistory:
+ def output(self):
+ return [{"ai": False, "content": "hello"}]
+
+
+class _CompactionLog:
+ def __init__(self):
+ self.logs = []
+ self.entries = []
+ self.reset_called = False
+ self.progress = None
+
+ def log(self, **kwargs):
+ self.entries.append(kwargs)
+ return _FakeLog()
+
+ def reset(self):
+ self.reset_called = True
+
+ def set_progress(self, *args, **kwargs):
+ self.progress = (args, kwargs)
+
+
+class _CompactionAgent:
+ DATA_NAME_RESPONSES_STATE = "responses_state"
+
+ def __init__(self):
+ self.history = _CompactionHistory()
+ self.data = {
+ "responses_state": {
+ "response_id": "resp_current",
+ "previous_response_id": "resp_previous",
+ "response_ids": ["resp_previous", "resp_current"],
+ }
+ }
+
+ def get_data(self, key):
+ return self.data.get(key)
+
+ def set_data(self, key, value):
+ self.data[key] = value
+
+
def test_pre_compaction_backup_sanitizes_surrogate_text(tmp_path, monkeypatch):
monkeypatch.setattr(
compactor,
@@ -114,3 +157,39 @@ async def test_large_compaction_does_not_send_unsplit_single_line_payload(monkey
assert len(chunk_messages) > 2
assert all(chunk_messages)
assert all(len(message) <= 10_000 for message in chunk_messages)
+
+
+@pytest.mark.asyncio
+async def test_manual_compaction_clears_active_responses_state(monkeypatch):
+ async def fake_single_pass(*args, **kwargs):
+ return "summary"
+
+ agent = _CompactionAgent()
+ context = SimpleNamespace(
+ id="compact-chat",
+ agent0=agent,
+ log=_CompactionLog(),
+ streaming_agent=object(),
+ )
+
+ monkeypatch.setattr(
+ compactor,
+ "_build_model",
+ lambda *args: ({"ctx_length": 128000}, _RecordingModel()),
+ )
+ monkeypatch.setattr(compactor, "_compact_single_pass", fake_single_pass)
+ monkeypatch.setattr(
+ compactor,
+ "_save_pre_compaction_backup",
+ lambda *args: {"txt": "/tmp/pre.txt"},
+ )
+ monkeypatch.setattr(compactor, "save_tmp_chat", lambda *args: None)
+ monkeypatch.setattr(compactor, "remove_msg_files", lambda *args: None)
+ monkeypatch.setattr(compactor, "mark_dirty_all", lambda *args, **kwargs: None)
+
+ await compactor.run_compaction(context)
+
+ state = agent.data["responses_state"]
+ assert "response_id" not in state
+ assert "previous_response_id" not in state
+ assert state["response_ids"] == ["resp_previous", "resp_current"]
diff --git a/tests/test_history_compression_wait.py b/tests/test_history_compression_wait.py
index ad6e81049..7e399b84f 100644
--- a/tests/test_history_compression_wait.py
+++ b/tests/test_history_compression_wait.py
@@ -45,6 +45,23 @@ class _MaxPassHistory:
return True
+class _CompressOnceHistory:
+ def __init__(self):
+ self.compress_calls = 0
+ self.tokens = 2000
+
+ def is_over_limit(self):
+ return self.compress_calls == 0
+
+ def get_tokens(self):
+ return self.tokens
+
+ async def compress(self):
+ self.compress_calls += 1
+ self.tokens -= 1000
+ return True
+
+
class _FakeLog:
def __init__(self):
self.entries = []
@@ -94,3 +111,20 @@ async def test_history_wait_stops_after_max_sync_compression_passes():
f"stopped after {MAX_SYNC_COMPRESSION_PASSES} passes"
in agent.context.log.entries[-1]["content"]
)
+
+
+@pytest.mark.asyncio
+async def test_history_compression_clears_active_responses_state():
+ agent = _FakeAgent(_CompressOnceHistory())
+ agent.data["responses_state"] = {
+ "response_id": "resp_current",
+ "previous_response_id": "resp_previous",
+ "response_ids": ["resp_previous", "resp_current"],
+ }
+
+ await OrganizeHistoryWait(agent).execute()
+
+ state = agent.data["responses_state"]
+ assert "response_id" not in state
+ assert "previous_response_id" not in state
+ assert state["response_ids"] == ["resp_previous", "resp_current"]
From 7aba42d46eb29f1095070299a8d3067fd7e14a45 Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Tue, 7 Jul 2026 17:01:46 +0200
Subject: [PATCH 22/43] Default local providers to chat completions
Force chat-completions transport by default for local and broad OpenAI-compatible chat providers whose Responses API implementations are often missing or unstable.\n\nCovers LM Studio, llama.cpp, Ollama, Ollama Cloud, oMLX, vLLM, and Other OpenAI compatible while leaving embedding provider defaults untouched. Updates provider-default regression coverage for the new chat-only defaults.
---
conf/model_providers.yaml | 8 ++++++++
tests/test_model_config_api_keys.py | 31 +++++++++++++++++++++++++++++
2 files changed, 39 insertions(+)
diff --git a/conf/model_providers.yaml b/conf/model_providers.yaml
index ba3f0ea48..58bb2e7cc 100644
--- a/conf/model_providers.yaml
+++ b/conf/model_providers.yaml
@@ -84,6 +84,7 @@ chat:
endpoint_url: "/v1/models"
default_base: "http://host.docker.internal:1234"
kwargs:
+ a0_api_mode: chat
api_base: "http://host.docker.internal:1234/v1"
api_key: "lm-studio"
llama_cpp:
@@ -93,6 +94,7 @@ chat:
endpoint_url: "/v1/models"
default_base: "http://host.docker.internal:8080"
kwargs:
+ a0_api_mode: chat
api_base: "http://host.docker.internal:8080/v1"
api_key: "llama-cpp"
mistral:
@@ -120,6 +122,7 @@ chat:
format: "ollama"
default_base: "http://host.docker.internal:11434"
kwargs:
+ a0_api_mode: chat
api_base: "http://host.docker.internal:11434"
omlx:
name: oMLX
@@ -128,6 +131,7 @@ chat:
endpoint_url: "/v1/models"
default_base: "http://host.docker.internal:8000"
kwargs:
+ a0_api_mode: chat
api_base: "http://host.docker.internal:8000/v1"
api_key: "omlx"
ollama_cloud:
@@ -136,6 +140,7 @@ chat:
models_list:
endpoint_url: "/models"
kwargs:
+ a0_api_mode: chat
api_base: https://ollama.com/v1
openai:
name: OpenAI
@@ -184,6 +189,7 @@ chat:
endpoint_url: "/v1/models"
default_base: "http://host.docker.internal:8000"
kwargs:
+ a0_api_mode: chat
api_base: "http://host.docker.internal:8000/v1"
api_key: "vllm"
xai:
@@ -208,6 +214,8 @@ chat:
other:
name: Other OpenAI compatible
litellm_provider: openai
+ kwargs:
+ a0_api_mode: chat
embedding:
huggingface:
diff --git a/tests/test_model_config_api_keys.py b/tests/test_model_config_api_keys.py
index e1f62d2f0..a76c1cd0d 100644
--- a/tests/test_model_config_api_keys.py
+++ b/tests/test_model_config_api_keys.py
@@ -265,6 +265,7 @@ def test_ollama_cloud_provider_config_requires_key_and_base_url():
ollama_cloud = provider_config["chat"]["ollama_cloud"]
assert ollama_cloud["name"] == "Ollama Cloud"
+ assert ollama_cloud["kwargs"]["a0_api_mode"] == "chat"
assert ollama_cloud["kwargs"]["api_base"] == "https://ollama.com/v1"
assert ollama_cloud["models_list"]["endpoint_url"] == "/models"
assert "api_key_mode" not in ollama_cloud
@@ -298,6 +299,36 @@ def test_direct_venice_chat_provider_defaults_to_chat_completions(monkeypatch):
assert custom.kwargs["a0_api_mode"] == "responses"
+def test_local_chat_providers_default_to_chat_completions():
+ import yaml
+
+ provider_path = PROJECT_ROOT / "conf/model_providers.yaml"
+ provider_config = yaml.safe_load(provider_path.read_text(encoding="utf-8"))
+
+ chat_completions_default_providers = (
+ "lm_studio",
+ "llama_cpp",
+ "ollama",
+ "ollama_cloud",
+ "omlx",
+ "other",
+ "vllm",
+ )
+ local_embedding_providers = (
+ "lm_studio",
+ "llama_cpp",
+ "ollama",
+ "omlx",
+ "vllm",
+ )
+
+ for provider in chat_completions_default_providers:
+ assert provider_config["chat"][provider]["kwargs"]["a0_api_mode"] == "chat"
+
+ for provider in local_embedding_providers:
+ assert "a0_api_mode" not in provider_config["embedding"][provider]["kwargs"]
+
+
def test_missing_api_key_banner_does_not_include_auto_modal_metadata(monkeypatch):
from plugins._model_config.helpers import model_config
From 5ac4f75a470ed0353e4b2f8d55002238ad64c066 Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Wed, 8 Jul 2026 01:28:36 +0200
Subject: [PATCH 23/43] Stop WebUI startup scripts from blocking render
Load startup vendor scripts with defer and serve Bootstrap locally so the first WebUI mount no longer depends on a parser-blocking CDN request.
Add a focused regression that rejects undeferred classic startup scripts in index.html and documents the vendor/startup contract.
---
tests/test_webui_startup_assets.py | 40 +++++++++++++++++++
webui/AGENTS.md | 1 +
webui/index.html | 12 ++----
webui/vendor/AGENTS.md | 1 +
webui/vendor/bootstrap/AGENTS.md | 27 +++++++++++++
.../vendor/bootstrap/bootstrap.bundle.min.js | 7 ++++
6 files changed, 80 insertions(+), 8 deletions(-)
create mode 100644 tests/test_webui_startup_assets.py
create mode 100644 webui/vendor/bootstrap/AGENTS.md
create mode 100644 webui/vendor/bootstrap/bootstrap.bundle.min.js
diff --git a/tests/test_webui_startup_assets.py b/tests/test_webui_startup_assets.py
new file mode 100644
index 000000000..93e7b4977
--- /dev/null
+++ b/tests/test_webui_startup_assets.py
@@ -0,0 +1,40 @@
+from pathlib import Path
+from html.parser import HTMLParser
+
+
+PROJECT_ROOT = Path(__file__).resolve().parents[1]
+
+
+class ScriptParser(HTMLParser):
+ def __init__(self) -> None:
+ super().__init__()
+ self.scripts: list[dict[str, str | None]] = []
+
+ def handle_starttag(self, tag: str, attrs: list[tuple[str, str | None]]) -> None:
+ if tag == "script":
+ self.scripts.append(dict(attrs))
+
+
+def test_bootstrap_is_local_and_deferred() -> None:
+ index_html = (PROJECT_ROOT / "webui" / "index.html").read_text(encoding="utf-8")
+
+ assert "cdn.jsdelivr.net/npm/bootstrap" not in index_html
+ assert '' in index_html
+ assert (PROJECT_ROOT / "webui" / "vendor" / "bootstrap" / "bootstrap.bundle.min.js").is_file()
+
+
+def test_classic_startup_scripts_are_deferred() -> None:
+ index_html = (PROJECT_ROOT / "webui" / "index.html").read_text(encoding="utf-8")
+ parser = ScriptParser()
+ parser.feed(index_html)
+
+ blocking_scripts = [
+ script["src"]
+ for script in parser.scripts
+ if script.get("src")
+ and script.get("type") != "module"
+ and "defer" not in script
+ and "async" not in script
+ ]
+
+ assert blocking_scripts == []
diff --git a/webui/AGENTS.md b/webui/AGENTS.md
index b0400b627..628a3a17b 100644
--- a/webui/AGENTS.md
+++ b/webui/AGENTS.md
@@ -25,6 +25,7 @@
- Component HTML loaded by the shared loader may include `
`, module scripts, body content, and scoped styles; modal content uses the same loader path.
- Do not bypass WebSocket origin/auth/CSRF assumptions from frontend code.
- Avoid editing vendored files unless intentionally updating the vendor asset.
+- Startup scripts in `index.html` must be local and non-parser-blocking: use ES modules, `defer`, or `async` for every script with `src`.
- Rubik (`--font-family-main`) is the default WebUI text and control font; use the code/mono font tokens only for code, logs, paths, and fixed-width data.
- Hover, focus, and active border treatments should follow existing neutral border/background patterns; avoid hard-coded blue border highlights unless matching an established specialized surface.
diff --git a/webui/index.html b/webui/index.html
index d7a2ee70c..2bb97e51b 100644
--- a/webui/index.html
+++ b/webui/index.html
@@ -23,7 +23,7 @@
<!-- Flatpickr for datetime picker -->
<link rel="stylesheet" href="vendor/flatpickr/flatpickr.min.css">
- <script src="vendor/flatpickr/flatpickr.min.js"></script>
+ <script defer src="vendor/flatpickr/flatpickr.min.js"></script>
<script>
globalThis.safeCall = function (name, ...args) {
@@ -34,15 +34,14 @@
<script type="module" src="index.js"></script>
<!-- Bootstrap JS (only for logic, importing bundled CSS => UI conflicts) -->
- <script src="https://cdn.jsdelivr.net/npm/bootstrap@5.0.2/dist/js/bootstrap.bundle.min.js"
- integrity="sha384-MrcW6ZMFYlzcLA8Nl+NtUVF0sA7MsXsP1UyJoMp4YLEuNSfAP+JcXn/tWtIaxVXM"
- crossorigin="anonymous"></script>
+ <script defer src="vendor/bootstrap/bootstrap.bundle.min.js"></script>
<!-- Then load Alpine.js -->
<!-- <script defer src="https://cdn.jsdelivr.net/npm/alpinejs@3.14.3/dist/cdn.min.js"></script> -->
+ <script defer src="vendor/ace-min/ace.js"></script>
+ <script defer src="vendor/qrcode.min.js"></script>
<script type="module" src="js/initFw.js"></script>
- <script src="vendor/ace-min/ace.js"></script>
<link href="vendor/ace-min/ace.min.css" rel="stylesheet">
<!-- KaTeX CSS -->
<link rel="stylesheet" href="vendor/katex/katex.min.css" crossorigin="anonymous">
@@ -51,9 +50,6 @@
<script defer src="vendor/katex/katex.min.js" crossorigin="anonymous"></script>
<script defer src="vendor/katex/katex.auto-render.min.js" crossorigin="anonymous"></script>
- <!-- QR Code Library -->
- <script src="vendor/qrcode.min.js"></script>
-
<!-- Google Icons -->
<link rel="stylesheet" href="vendor/google/google-icons.css" />
diff --git a/webui/vendor/AGENTS.md b/webui/vendor/AGENTS.md
index 28ad931ce..446f363d7 100644
--- a/webui/vendor/AGENTS.md
+++ b/webui/vendor/AGENTS.md
@@ -35,6 +35,7 @@ Direct child DOX files:
| [_ace/AGENTS.md](_ace/AGENTS.md) | Minimal Ace editor subset used by core editor surfaces. |
| [ace-min/AGENTS.md](ace-min/AGENTS.md) | Full minified Ace distribution and supporting assets. |
| [alpine/AGENTS.md](alpine/AGENTS.md) | Vendored Alpine.js core and collapse plugin. |
+| [bootstrap/AGENTS.md](bootstrap/AGENTS.md) | Vendored Bootstrap JavaScript bundle for collapse and tooltip behavior. |
| [dompurify/AGENTS.md](dompurify/AGENTS.md) | Vendored DOMPurify sanitizer module. |
| [flatpickr/AGENTS.md](flatpickr/AGENTS.md) | Vendored Flatpickr date/time picker assets. |
| [google/AGENTS.md](google/AGENTS.md) | Vendored Google Material Symbols font and stylesheet. |
diff --git a/webui/vendor/bootstrap/AGENTS.md b/webui/vendor/bootstrap/AGENTS.md
new file mode 100644
index 000000000..57af1493e
--- /dev/null
+++ b/webui/vendor/bootstrap/AGENTS.md
@@ -0,0 +1,27 @@
+# Bootstrap Vendor DOX
+
+## Purpose
+
+- Own the vendored Bootstrap JavaScript bundle used for collapse and tooltip behavior.
+
+## Ownership
+
+- `bootstrap.bundle.min.js` owns Bootstrap runtime behavior.
+
+## Local Contracts
+
+- Do not hand-edit the minified vendor file for application behavior.
+- Keep `webui/index.html` synchronized with the local bundle path.
+- Bootstrap must be served locally and loaded with `defer` so startup does not wait on an external parser-blocking script.
+
+## Work Guidance
+
+- Replace from a clean upstream release when updating.
+
+## Verification
+
+- Smoke-test sidebar collapse controls and tooltips after changes.
+
+## Child DOX Index
+
+No child DOX files.
diff --git a/webui/vendor/bootstrap/bootstrap.bundle.min.js b/webui/vendor/bootstrap/bootstrap.bundle.min.js
new file mode 100644
index 000000000..68acb7a31
--- /dev/null
+++ b/webui/vendor/bootstrap/bootstrap.bundle.min.js
@@ -0,0 +1,7 @@
+/*!
+ * Bootstrap v5.0.2 (https://getbootstrap.com/)
+ * Copyright 2011-2021 The Bootstrap Authors (https://github.com/twbs/bootstrap/graphs/contributors)
+ * Licensed under MIT (https://github.com/twbs/bootstrap/blob/main/LICENSE)
+ */
+!function(t,e){"object"==typeof exports&&"undefined"!=typeof module?module.exports=e():"function"==typeof define&&define.amd?define(e):(t="undefined"!=typeof globalThis?globalThis:t||self).bootstrap=e()}(this,(function(){"use strict";const t={find:(t,e=document.documentElement)=>[].concat(...Element.prototype.querySelectorAll.call(e,t)),findOne:(t,e=document.documentElement)=>Element.prototype.querySelector.call(e,t),children:(t,e)=>[].concat(...t.children).filter(t=>t.matches(e)),parents(t,e){const i=[];let n=t.parentNode;for(;n&&n.nodeType===Node.ELEMENT_NODE&&3!==n.nodeType;)n.matches(e)&&i.push(n),n=n.parentNode;return i},prev(t,e){let i=t.previousElementSibling;for(;i;){if(i.matches(e))return[i];i=i.previousElementSibling}return[]},next(t,e){let i=t.nextElementSibling;for(;i;){if(i.matches(e))return[i];i=i.nextElementSibling}return[]}},e=t=>{do{t+=Math.floor(1e6*Math.random())}while(document.getElementById(t));return t},i=t=>{let e=t.getAttribute("data-bs-target");if(!e||"#"===e){let i=t.getAttribute("href");if(!i||!i.includes("#")&&!i.startsWith("."))return null;i.includes("#")&&!i.startsWith("#")&&(i="#"+i.split("#")[1]),e=i&&"#"!==i?i.trim():null}return e},n=t=>{const e=i(t);return e&&document.querySelector(e)?e:null},s=t=>{const e=i(t);return e?document.querySelector(e):null},o=t=>{t.dispatchEvent(new Event("transitionend"))},r=t=>!(!t||"object"!=typeof t)&&(void 0!==t.jquery&&(t=t[0]),void 0!==t.nodeType),a=e=>r(e)?e.jquery?e[0]:e:"string"==typeof e&&e.length>0?t.findOne(e):null,l=(t,e,i)=>{Object.keys(i).forEach(n=>{const s=i[n],o=e[n],a=o&&r(o)?"element":null==(l=o)?""+l:{}.toString.call(l).match(/\s([a-z]+)/i)[1].toLowerCase();var l;if(!new RegExp(s).test(a))throw new TypeError(`${t.toUpperCase()}: Option "${n}" provided type "${a}" but expected type "${s}".`)})},c=t=>!(!r(t)||0===t.getClientRects().length)&&"visible"===getComputedStyle(t).getPropertyValue("visibility"),h=t=>!t||t.nodeType!==Node.ELEMENT_NODE||!!t.classList.contains("disabled")||(void 0!==t.disabled?t.disabled:t.hasAttribute("disabled")&&"false"!==t.getAttribute("disabled")),d=t=>{if(!document.documentElement.attachShadow)return null;if("function"==typeof t.getRootNode){const e=t.getRootNode();return e instanceof ShadowRoot?e:null}return t instanceof ShadowRoot?t:t.parentNode?d(t.parentNode):null},u=()=>{},f=t=>t.offsetHeight,p=()=>{const{jQuery:t}=window;return t&&!document.body.hasAttribute("data-bs-no-jquery")?t:null},m=[],g=()=>"rtl"===document.documentElement.dir,_=t=>{var e;e=()=>{const e=p();if(e){const i=t.NAME,n=e.fn[i];e.fn[i]=t.jQueryInterface,e.fn[i].Constructor=t,e.fn[i].noConflict=()=>(e.fn[i]=n,t.jQueryInterface)}},"loading"===document.readyState?(m.length||document.addEventListener("DOMContentLoaded",()=>{m.forEach(t=>t())}),m.push(e)):e()},b=t=>{"function"==typeof t&&t()},v=(t,e,i=!0)=>{if(!i)return void b(t);const n=(t=>{if(!t)return 0;let{transitionDuration:e,transitionDelay:i}=window.getComputedStyle(t);const n=Number.parseFloat(e),s=Number.parseFloat(i);return n||s?(e=e.split(",")[0],i=i.split(",")[0],1e3*(Number.parseFloat(e)+Number.parseFloat(i))):0})(e)+5;let s=!1;const r=({target:i})=>{i===e&&(s=!0,e.removeEventListener("transitionend",r),b(t))};e.addEventListener("transitionend",r),setTimeout(()=>{s||o(e)},n)},y=(t,e,i,n)=>{let s=t.indexOf(e);if(-1===s)return t[!i&&n?t.length-1:0];const o=t.length;return s+=i?1:-1,n&&(s=(s+o)%o),t[Math.max(0,Math.min(s,o-1))]},w=/[^.]*(?=\..*)\.|.*/,E=/\..*/,A=/::\d+$/,T={};let O=1;const C={mouseenter:"mouseover",mouseleave:"mouseout"},k=/^(mouseenter|mouseleave)/i,L=new Set(["click","dblclick","mouseup","mousedown","contextmenu","mousewheel","DOMMouseScroll","mouseover","mouseout","mousemove","selectstart","selectend","keydown","keypress","keyup","orientationchange","touchstart","touchmove","touchend","touchcancel","pointerdown","pointermove","pointerup","pointerleave","pointercancel","gesturestart","gesturechange","gestureend","focus","blur","change","reset","select","submit","focusin","focusout","load","unload","beforeunload","resize","move","DOMContentLoaded","readystatechange","error","abort","scroll"]);function x(t,e){return e&&`${e}::${O++}`||t.uidEvent||O++}function D(t){const e=x(t);return t.uidEvent=e,T[e]=T[e]||{},T[e]}function S(t,e,i=null){const n=Object.keys(t);for(let s=0,o=n.length;s<o;s++){const o=t[n[s]];if(o.originalHandler===e&&o.delegationSelector===i)return o}return null}function I(t,e,i){const n="string"==typeof e,s=n?i:e;let o=M(t);return L.has(o)||(o=t),[n,s,o]}function N(t,e,i,n,s){if("string"!=typeof e||!t)return;if(i||(i=n,n=null),k.test(e)){const t=t=>function(e){if(!e.relatedTarget||e.relatedTarget!==e.delegateTarget&&!e.delegateTarget.contains(e.relatedTarget))return t.call(this,e)};n?n=t(n):i=t(i)}const[o,r,a]=I(e,i,n),l=D(t),c=l[a]||(l[a]={}),h=S(c,r,o?i:null);if(h)return void(h.oneOff=h.oneOff&&s);const d=x(r,e.replace(w,"")),u=o?function(t,e,i){return function n(s){const o=t.querySelectorAll(e);for(let{target:r}=s;r&&r!==this;r=r.parentNode)for(let a=o.length;a--;)if(o[a]===r)return s.delegateTarget=r,n.oneOff&&P.off(t,s.type,e,i),i.apply(r,[s]);return null}}(t,i,n):function(t,e){return function i(n){return n.delegateTarget=t,i.oneOff&&P.off(t,n.type,e),e.apply(t,[n])}}(t,i);u.delegationSelector=o?i:null,u.originalHandler=r,u.oneOff=s,u.uidEvent=d,c[d]=u,t.addEventListener(a,u,o)}function j(t,e,i,n,s){const o=S(e[i],n,s);o&&(t.removeEventListener(i,o,Boolean(s)),delete e[i][o.uidEvent])}function M(t){return t=t.replace(E,""),C[t]||t}const P={on(t,e,i,n){N(t,e,i,n,!1)},one(t,e,i,n){N(t,e,i,n,!0)},off(t,e,i,n){if("string"!=typeof e||!t)return;const[s,o,r]=I(e,i,n),a=r!==e,l=D(t),c=e.startsWith(".");if(void 0!==o){if(!l||!l[r])return;return void j(t,l,r,o,s?i:null)}c&&Object.keys(l).forEach(i=>{!function(t,e,i,n){const s=e[i]||{};Object.keys(s).forEach(o=>{if(o.includes(n)){const n=s[o];j(t,e,i,n.originalHandler,n.delegationSelector)}})}(t,l,i,e.slice(1))});const h=l[r]||{};Object.keys(h).forEach(i=>{const n=i.replace(A,"");if(!a||e.includes(n)){const e=h[i];j(t,l,r,e.originalHandler,e.delegationSelector)}})},trigger(t,e,i){if("string"!=typeof e||!t)return null;const n=p(),s=M(e),o=e!==s,r=L.has(s);let a,l=!0,c=!0,h=!1,d=null;return o&&n&&(a=n.Event(e,i),n(t).trigger(a),l=!a.isPropagationStopped(),c=!a.isImmediatePropagationStopped(),h=a.isDefaultPrevented()),r?(d=document.createEvent("HTMLEvents"),d.initEvent(s,l,!0)):d=new CustomEvent(e,{bubbles:l,cancelable:!0}),void 0!==i&&Object.keys(i).forEach(t=>{Object.defineProperty(d,t,{get:()=>i[t]})}),h&&d.preventDefault(),c&&t.dispatchEvent(d),d.defaultPrevented&&void 0!==a&&a.preventDefault(),d}},H=new Map;var R={set(t,e,i){H.has(t)||H.set(t,new Map);const n=H.get(t);n.has(e)||0===n.size?n.set(e,i):console.error(`Bootstrap doesn't allow more than one instance per element. Bound instance: ${Array.from(n.keys())[0]}.`)},get:(t,e)=>H.has(t)&&H.get(t).get(e)||null,remove(t,e){if(!H.has(t))return;const i=H.get(t);i.delete(e),0===i.size&&H.delete(t)}};class B{constructor(t){(t=a(t))&&(this._element=t,R.set(this._element,this.constructor.DATA_KEY,this))}dispose(){R.remove(this._element,this.constructor.DATA_KEY),P.off(this._element,this.constructor.EVENT_KEY),Object.getOwnPropertyNames(this).forEach(t=>{this[t]=null})}_queueCallback(t,e,i=!0){v(t,e,i)}static getInstance(t){return R.get(t,this.DATA_KEY)}static getOrCreateInstance(t,e={}){return this.getInstance(t)||new this(t,"object"==typeof e?e:null)}static get VERSION(){return"5.0.2"}static get NAME(){throw new Error('You have to implement the static method "NAME", for each component!')}static get DATA_KEY(){return"bs."+this.NAME}static get EVENT_KEY(){return"."+this.DATA_KEY}}class W extends B{static get NAME(){return"alert"}close(t){const e=t?this._getRootElement(t):this._element,i=this._triggerCloseEvent(e);null===i||i.defaultPrevented||this._removeElement(e)}_getRootElement(t){return s(t)||t.closest(".alert")}_triggerCloseEvent(t){return P.trigger(t,"close.bs.alert")}_removeElement(t){t.classList.remove("show");const e=t.classList.contains("fade");this._queueCallback(()=>this._destroyElement(t),t,e)}_destroyElement(t){t.remove(),P.trigger(t,"closed.bs.alert")}static jQueryInterface(t){return this.each((function(){const e=W.getOrCreateInstance(this);"close"===t&&e[t](this)}))}static handleDismiss(t){return function(e){e&&e.preventDefault(),t.close(this)}}}P.on(document,"click.bs.alert.data-api",'[data-bs-dismiss="alert"]',W.handleDismiss(new W)),_(W);class q extends B{static get NAME(){return"button"}toggle(){this._element.setAttribute("aria-pressed",this._element.classList.toggle("active"))}static jQueryInterface(t){return this.each((function(){const e=q.getOrCreateInstance(this);"toggle"===t&&e[t]()}))}}function z(t){return"true"===t||"false"!==t&&(t===Number(t).toString()?Number(t):""===t||"null"===t?null:t)}function $(t){return t.replace(/[A-Z]/g,t=>"-"+t.toLowerCase())}P.on(document,"click.bs.button.data-api",'[data-bs-toggle="button"]',t=>{t.preventDefault();const e=t.target.closest('[data-bs-toggle="button"]');q.getOrCreateInstance(e).toggle()}),_(q);const U={setDataAttribute(t,e,i){t.setAttribute("data-bs-"+$(e),i)},removeDataAttribute(t,e){t.removeAttribute("data-bs-"+$(e))},getDataAttributes(t){if(!t)return{};const e={};return Object.keys(t.dataset).filter(t=>t.startsWith("bs")).forEach(i=>{let n=i.replace(/^bs/,"");n=n.charAt(0).toLowerCase()+n.slice(1,n.length),e[n]=z(t.dataset[i])}),e},getDataAttribute:(t,e)=>z(t.getAttribute("data-bs-"+$(e))),offset(t){const e=t.getBoundingClientRect();return{top:e.top+document.body.scrollTop,left:e.left+document.body.scrollLeft}},position:t=>({top:t.offsetTop,left:t.offsetLeft})},F={interval:5e3,keyboard:!0,slide:!1,pause:"hover",wrap:!0,touch:!0},V={interval:"(number|boolean)",keyboard:"boolean",slide:"(boolean|string)",pause:"(string|boolean)",wrap:"boolean",touch:"boolean"},K="next",X="prev",Y="left",Q="right",G={ArrowLeft:Q,ArrowRight:Y};class Z extends B{constructor(e,i){super(e),this._items=null,this._interval=null,this._activeElement=null,this._isPaused=!1,this._isSliding=!1,this.touchTimeout=null,this.touchStartX=0,this.touchDeltaX=0,this._config=this._getConfig(i),this._indicatorsElement=t.findOne(".carousel-indicators",this._element),this._touchSupported="ontouchstart"in document.documentElement||navigator.maxTouchPoints>0,this._pointerEvent=Boolean(window.PointerEvent),this._addEventListeners()}static get Default(){return F}static get NAME(){return"carousel"}next(){this._slide(K)}nextWhenVisible(){!document.hidden&&c(this._element)&&this.next()}prev(){this._slide(X)}pause(e){e||(this._isPaused=!0),t.findOne(".carousel-item-next, .carousel-item-prev",this._element)&&(o(this._element),this.cycle(!0)),clearInterval(this._interval),this._interval=null}cycle(t){t||(this._isPaused=!1),this._interval&&(clearInterval(this._interval),this._interval=null),this._config&&this._config.interval&&!this._isPaused&&(this._updateInterval(),this._interval=setInterval((document.visibilityState?this.nextWhenVisible:this.next).bind(this),this._config.interval))}to(e){this._activeElement=t.findOne(".active.carousel-item",this._element);const i=this._getItemIndex(this._activeElement);if(e>this._items.length-1||e<0)return;if(this._isSliding)return void P.one(this._element,"slid.bs.carousel",()=>this.to(e));if(i===e)return this.pause(),void this.cycle();const n=e>i?K:X;this._slide(n,this._items[e])}_getConfig(t){return t={...F,...U.getDataAttributes(this._element),..."object"==typeof t?t:{}},l("carousel",t,V),t}_handleSwipe(){const t=Math.abs(this.touchDeltaX);if(t<=40)return;const e=t/this.touchDeltaX;this.touchDeltaX=0,e&&this._slide(e>0?Q:Y)}_addEventListeners(){this._config.keyboard&&P.on(this._element,"keydown.bs.carousel",t=>this._keydown(t)),"hover"===this._config.pause&&(P.on(this._element,"mouseenter.bs.carousel",t=>this.pause(t)),P.on(this._element,"mouseleave.bs.carousel",t=>this.cycle(t))),this._config.touch&&this._touchSupported&&this._addTouchEventListeners()}_addTouchEventListeners(){const e=t=>{!this._pointerEvent||"pen"!==t.pointerType&&"touch"!==t.pointerType?this._pointerEvent||(this.touchStartX=t.touches[0].clientX):this.touchStartX=t.clientX},i=t=>{this.touchDeltaX=t.touches&&t.touches.length>1?0:t.touches[0].clientX-this.touchStartX},n=t=>{!this._pointerEvent||"pen"!==t.pointerType&&"touch"!==t.pointerType||(this.touchDeltaX=t.clientX-this.touchStartX),this._handleSwipe(),"hover"===this._config.pause&&(this.pause(),this.touchTimeout&&clearTimeout(this.touchTimeout),this.touchTimeout=setTimeout(t=>this.cycle(t),500+this._config.interval))};t.find(".carousel-item img",this._element).forEach(t=>{P.on(t,"dragstart.bs.carousel",t=>t.preventDefault())}),this._pointerEvent?(P.on(this._element,"pointerdown.bs.carousel",t=>e(t)),P.on(this._element,"pointerup.bs.carousel",t=>n(t)),this._element.classList.add("pointer-event")):(P.on(this._element,"touchstart.bs.carousel",t=>e(t)),P.on(this._element,"touchmove.bs.carousel",t=>i(t)),P.on(this._element,"touchend.bs.carousel",t=>n(t)))}_keydown(t){if(/input|textarea/i.test(t.target.tagName))return;const e=G[t.key];e&&(t.preventDefault(),this._slide(e))}_getItemIndex(e){return this._items=e&&e.parentNode?t.find(".carousel-item",e.parentNode):[],this._items.indexOf(e)}_getItemByOrder(t,e){const i=t===K;return y(this._items,e,i,this._config.wrap)}_triggerSlideEvent(e,i){const n=this._getItemIndex(e),s=this._getItemIndex(t.findOne(".active.carousel-item",this._element));return P.trigger(this._element,"slide.bs.carousel",{relatedTarget:e,direction:i,from:s,to:n})}_setActiveIndicatorElement(e){if(this._indicatorsElement){const i=t.findOne(".active",this._indicatorsElement);i.classList.remove("active"),i.removeAttribute("aria-current");const n=t.find("[data-bs-target]",this._indicatorsElement);for(let t=0;t<n.length;t++)if(Number.parseInt(n[t].getAttribute("data-bs-slide-to"),10)===this._getItemIndex(e)){n[t].classList.add("active"),n[t].setAttribute("aria-current","true");break}}}_updateInterval(){const e=this._activeElement||t.findOne(".active.carousel-item",this._element);if(!e)return;const i=Number.parseInt(e.getAttribute("data-bs-interval"),10);i?(this._config.defaultInterval=this._config.defaultInterval||this._config.interval,this._config.interval=i):this._config.interval=this._config.defaultInterval||this._config.interval}_slide(e,i){const n=this._directionToOrder(e),s=t.findOne(".active.carousel-item",this._element),o=this._getItemIndex(s),r=i||this._getItemByOrder(n,s),a=this._getItemIndex(r),l=Boolean(this._interval),c=n===K,h=c?"carousel-item-start":"carousel-item-end",d=c?"carousel-item-next":"carousel-item-prev",u=this._orderToDirection(n);if(r&&r.classList.contains("active"))return void(this._isSliding=!1);if(this._isSliding)return;if(this._triggerSlideEvent(r,u).defaultPrevented)return;if(!s||!r)return;this._isSliding=!0,l&&this.pause(),this._setActiveIndicatorElement(r),this._activeElement=r;const p=()=>{P.trigger(this._element,"slid.bs.carousel",{relatedTarget:r,direction:u,from:o,to:a})};if(this._element.classList.contains("slide")){r.classList.add(d),f(r),s.classList.add(h),r.classList.add(h);const t=()=>{r.classList.remove(h,d),r.classList.add("active"),s.classList.remove("active",d,h),this._isSliding=!1,setTimeout(p,0)};this._queueCallback(t,s,!0)}else s.classList.remove("active"),r.classList.add("active"),this._isSliding=!1,p();l&&this.cycle()}_directionToOrder(t){return[Q,Y].includes(t)?g()?t===Y?X:K:t===Y?K:X:t}_orderToDirection(t){return[K,X].includes(t)?g()?t===X?Y:Q:t===X?Q:Y:t}static carouselInterface(t,e){const i=Z.getOrCreateInstance(t,e);let{_config:n}=i;"object"==typeof e&&(n={...n,...e});const s="string"==typeof e?e:n.slide;if("number"==typeof e)i.to(e);else if("string"==typeof s){if(void 0===i[s])throw new TypeError(`No method named "${s}"`);i[s]()}else n.interval&&n.ride&&(i.pause(),i.cycle())}static jQueryInterface(t){return this.each((function(){Z.carouselInterface(this,t)}))}static dataApiClickHandler(t){const e=s(this);if(!e||!e.classList.contains("carousel"))return;const i={...U.getDataAttributes(e),...U.getDataAttributes(this)},n=this.getAttribute("data-bs-slide-to");n&&(i.interval=!1),Z.carouselInterface(e,i),n&&Z.getInstance(e).to(n),t.preventDefault()}}P.on(document,"click.bs.carousel.data-api","[data-bs-slide], [data-bs-slide-to]",Z.dataApiClickHandler),P.on(window,"load.bs.carousel.data-api",()=>{const e=t.find('[data-bs-ride="carousel"]');for(let t=0,i=e.length;t<i;t++)Z.carouselInterface(e[t],Z.getInstance(e[t]))}),_(Z);const J={toggle:!0,parent:""},tt={toggle:"boolean",parent:"(string|element)"};class et extends B{constructor(e,i){super(e),this._isTransitioning=!1,this._config=this._getConfig(i),this._triggerArray=t.find(`[data-bs-toggle="collapse"][href="#${this._element.id}"],[data-bs-toggle="collapse"][data-bs-target="#${this._element.id}"]`);const s=t.find('[data-bs-toggle="collapse"]');for(let e=0,i=s.length;e<i;e++){const i=s[e],o=n(i),r=t.find(o).filter(t=>t===this._element);null!==o&&r.length&&(this._selector=o,this._triggerArray.push(i))}this._parent=this._config.parent?this._getParent():null,this._config.parent||this._addAriaAndCollapsedClass(this._element,this._triggerArray),this._config.toggle&&this.toggle()}static get Default(){return J}static get NAME(){return"collapse"}toggle(){this._element.classList.contains("show")?this.hide():this.show()}show(){if(this._isTransitioning||this._element.classList.contains("show"))return;let e,i;this._parent&&(e=t.find(".show, .collapsing",this._parent).filter(t=>"string"==typeof this._config.parent?t.getAttribute("data-bs-parent")===this._config.parent:t.classList.contains("collapse")),0===e.length&&(e=null));const n=t.findOne(this._selector);if(e){const t=e.find(t=>n!==t);if(i=t?et.getInstance(t):null,i&&i._isTransitioning)return}if(P.trigger(this._element,"show.bs.collapse").defaultPrevented)return;e&&e.forEach(t=>{n!==t&&et.collapseInterface(t,"hide"),i||R.set(t,"bs.collapse",null)});const s=this._getDimension();this._element.classList.remove("collapse"),this._element.classList.add("collapsing"),this._element.style[s]=0,this._triggerArray.length&&this._triggerArray.forEach(t=>{t.classList.remove("collapsed"),t.setAttribute("aria-expanded",!0)}),this.setTransitioning(!0);const o="scroll"+(s[0].toUpperCase()+s.slice(1));this._queueCallback(()=>{this._element.classList.remove("collapsing"),this._element.classList.add("collapse","show"),this._element.style[s]="",this.setTransitioning(!1),P.trigger(this._element,"shown.bs.collapse")},this._element,!0),this._element.style[s]=this._element[o]+"px"}hide(){if(this._isTransitioning||!this._element.classList.contains("show"))return;if(P.trigger(this._element,"hide.bs.collapse").defaultPrevented)return;const t=this._getDimension();this._element.style[t]=this._element.getBoundingClientRect()[t]+"px",f(this._element),this._element.classList.add("collapsing"),this._element.classList.remove("collapse","show");const e=this._triggerArray.length;if(e>0)for(let t=0;t<e;t++){const e=this._triggerArray[t],i=s(e);i&&!i.classList.contains("show")&&(e.classList.add("collapsed"),e.setAttribute("aria-expanded",!1))}this.setTransitioning(!0),this._element.style[t]="",this._queueCallback(()=>{this.setTransitioning(!1),this._element.classList.remove("collapsing"),this._element.classList.add("collapse"),P.trigger(this._element,"hidden.bs.collapse")},this._element,!0)}setTransitioning(t){this._isTransitioning=t}_getConfig(t){return(t={...J,...t}).toggle=Boolean(t.toggle),l("collapse",t,tt),t}_getDimension(){return this._element.classList.contains("width")?"width":"height"}_getParent(){let{parent:e}=this._config;e=a(e);const i=`[data-bs-toggle="collapse"][data-bs-parent="${e}"]`;return t.find(i,e).forEach(t=>{const e=s(t);this._addAriaAndCollapsedClass(e,[t])}),e}_addAriaAndCollapsedClass(t,e){if(!t||!e.length)return;const i=t.classList.contains("show");e.forEach(t=>{i?t.classList.remove("collapsed"):t.classList.add("collapsed"),t.setAttribute("aria-expanded",i)})}static collapseInterface(t,e){let i=et.getInstance(t);const n={...J,...U.getDataAttributes(t),..."object"==typeof e&&e?e:{}};if(!i&&n.toggle&&"string"==typeof e&&/show|hide/.test(e)&&(n.toggle=!1),i||(i=new et(t,n)),"string"==typeof e){if(void 0===i[e])throw new TypeError(`No method named "${e}"`);i[e]()}}static jQueryInterface(t){return this.each((function(){et.collapseInterface(this,t)}))}}P.on(document,"click.bs.collapse.data-api",'[data-bs-toggle="collapse"]',(function(e){("A"===e.target.tagName||e.delegateTarget&&"A"===e.delegateTarget.tagName)&&e.preventDefault();const i=U.getDataAttributes(this),s=n(this);t.find(s).forEach(t=>{const e=et.getInstance(t);let n;e?(null===e._parent&&"string"==typeof i.parent&&(e._config.parent=i.parent,e._parent=e._getParent()),n="toggle"):n=i,et.collapseInterface(t,n)})})),_(et);var it="top",nt="bottom",st="right",ot="left",rt=[it,nt,st,ot],at=rt.reduce((function(t,e){return t.concat([e+"-start",e+"-end"])}),[]),lt=[].concat(rt,["auto"]).reduce((function(t,e){return t.concat([e,e+"-start",e+"-end"])}),[]),ct=["beforeRead","read","afterRead","beforeMain","main","afterMain","beforeWrite","write","afterWrite"];function ht(t){return t?(t.nodeName||"").toLowerCase():null}function dt(t){if(null==t)return window;if("[object Window]"!==t.toString()){var e=t.ownerDocument;return e&&e.defaultView||window}return t}function ut(t){return t instanceof dt(t).Element||t instanceof Element}function ft(t){return t instanceof dt(t).HTMLElement||t instanceof HTMLElement}function pt(t){return"undefined"!=typeof ShadowRoot&&(t instanceof dt(t).ShadowRoot||t instanceof ShadowRoot)}var mt={name:"applyStyles",enabled:!0,phase:"write",fn:function(t){var e=t.state;Object.keys(e.elements).forEach((function(t){var i=e.styles[t]||{},n=e.attributes[t]||{},s=e.elements[t];ft(s)&&ht(s)&&(Object.assign(s.style,i),Object.keys(n).forEach((function(t){var e=n[t];!1===e?s.removeAttribute(t):s.setAttribute(t,!0===e?"":e)})))}))},effect:function(t){var e=t.state,i={popper:{position:e.options.strategy,left:"0",top:"0",margin:"0"},arrow:{position:"absolute"},reference:{}};return Object.assign(e.elements.popper.style,i.popper),e.styles=i,e.elements.arrow&&Object.assign(e.elements.arrow.style,i.arrow),function(){Object.keys(e.elements).forEach((function(t){var n=e.elements[t],s=e.attributes[t]||{},o=Object.keys(e.styles.hasOwnProperty(t)?e.styles[t]:i[t]).reduce((function(t,e){return t[e]="",t}),{});ft(n)&&ht(n)&&(Object.assign(n.style,o),Object.keys(s).forEach((function(t){n.removeAttribute(t)})))}))}},requires:["computeStyles"]};function gt(t){return t.split("-")[0]}function _t(t){var e=t.getBoundingClientRect();return{width:e.width,height:e.height,top:e.top,right:e.right,bottom:e.bottom,left:e.left,x:e.left,y:e.top}}function bt(t){var e=_t(t),i=t.offsetWidth,n=t.offsetHeight;return Math.abs(e.width-i)<=1&&(i=e.width),Math.abs(e.height-n)<=1&&(n=e.height),{x:t.offsetLeft,y:t.offsetTop,width:i,height:n}}function vt(t,e){var i=e.getRootNode&&e.getRootNode();if(t.contains(e))return!0;if(i&&pt(i)){var n=e;do{if(n&&t.isSameNode(n))return!0;n=n.parentNode||n.host}while(n)}return!1}function yt(t){return dt(t).getComputedStyle(t)}function wt(t){return["table","td","th"].indexOf(ht(t))>=0}function Et(t){return((ut(t)?t.ownerDocument:t.document)||window.document).documentElement}function At(t){return"html"===ht(t)?t:t.assignedSlot||t.parentNode||(pt(t)?t.host:null)||Et(t)}function Tt(t){return ft(t)&&"fixed"!==yt(t).position?t.offsetParent:null}function Ot(t){for(var e=dt(t),i=Tt(t);i&&wt(i)&&"static"===yt(i).position;)i=Tt(i);return i&&("html"===ht(i)||"body"===ht(i)&&"static"===yt(i).position)?e:i||function(t){var e=-1!==navigator.userAgent.toLowerCase().indexOf("firefox");if(-1!==navigator.userAgent.indexOf("Trident")&&ft(t)&&"fixed"===yt(t).position)return null;for(var i=At(t);ft(i)&&["html","body"].indexOf(ht(i))<0;){var n=yt(i);if("none"!==n.transform||"none"!==n.perspective||"paint"===n.contain||-1!==["transform","perspective"].indexOf(n.willChange)||e&&"filter"===n.willChange||e&&n.filter&&"none"!==n.filter)return i;i=i.parentNode}return null}(t)||e}function Ct(t){return["top","bottom"].indexOf(t)>=0?"x":"y"}var kt=Math.max,Lt=Math.min,xt=Math.round;function Dt(t,e,i){return kt(t,Lt(e,i))}function St(t){return Object.assign({},{top:0,right:0,bottom:0,left:0},t)}function It(t,e){return e.reduce((function(e,i){return e[i]=t,e}),{})}var Nt={name:"arrow",enabled:!0,phase:"main",fn:function(t){var e,i=t.state,n=t.name,s=t.options,o=i.elements.arrow,r=i.modifiersData.popperOffsets,a=gt(i.placement),l=Ct(a),c=[ot,st].indexOf(a)>=0?"height":"width";if(o&&r){var h=function(t,e){return St("number"!=typeof(t="function"==typeof t?t(Object.assign({},e.rects,{placement:e.placement})):t)?t:It(t,rt))}(s.padding,i),d=bt(o),u="y"===l?it:ot,f="y"===l?nt:st,p=i.rects.reference[c]+i.rects.reference[l]-r[l]-i.rects.popper[c],m=r[l]-i.rects.reference[l],g=Ot(o),_=g?"y"===l?g.clientHeight||0:g.clientWidth||0:0,b=p/2-m/2,v=h[u],y=_-d[c]-h[f],w=_/2-d[c]/2+b,E=Dt(v,w,y),A=l;i.modifiersData[n]=((e={})[A]=E,e.centerOffset=E-w,e)}},effect:function(t){var e=t.state,i=t.options.element,n=void 0===i?"[data-popper-arrow]":i;null!=n&&("string"!=typeof n||(n=e.elements.popper.querySelector(n)))&&vt(e.elements.popper,n)&&(e.elements.arrow=n)},requires:["popperOffsets"],requiresIfExists:["preventOverflow"]},jt={top:"auto",right:"auto",bottom:"auto",left:"auto"};function Mt(t){var e,i=t.popper,n=t.popperRect,s=t.placement,o=t.offsets,r=t.position,a=t.gpuAcceleration,l=t.adaptive,c=t.roundOffsets,h=!0===c?function(t){var e=t.x,i=t.y,n=window.devicePixelRatio||1;return{x:xt(xt(e*n)/n)||0,y:xt(xt(i*n)/n)||0}}(o):"function"==typeof c?c(o):o,d=h.x,u=void 0===d?0:d,f=h.y,p=void 0===f?0:f,m=o.hasOwnProperty("x"),g=o.hasOwnProperty("y"),_=ot,b=it,v=window;if(l){var y=Ot(i),w="clientHeight",E="clientWidth";y===dt(i)&&"static"!==yt(y=Et(i)).position&&(w="scrollHeight",E="scrollWidth"),y=y,s===it&&(b=nt,p-=y[w]-n.height,p*=a?1:-1),s===ot&&(_=st,u-=y[E]-n.width,u*=a?1:-1)}var A,T=Object.assign({position:r},l&&jt);return a?Object.assign({},T,((A={})[b]=g?"0":"",A[_]=m?"0":"",A.transform=(v.devicePixelRatio||1)<2?"translate("+u+"px, "+p+"px)":"translate3d("+u+"px, "+p+"px, 0)",A)):Object.assign({},T,((e={})[b]=g?p+"px":"",e[_]=m?u+"px":"",e.transform="",e))}var Pt={name:"computeStyles",enabled:!0,phase:"beforeWrite",fn:function(t){var e=t.state,i=t.options,n=i.gpuAcceleration,s=void 0===n||n,o=i.adaptive,r=void 0===o||o,a=i.roundOffsets,l=void 0===a||a,c={placement:gt(e.placement),popper:e.elements.popper,popperRect:e.rects.popper,gpuAcceleration:s};null!=e.modifiersData.popperOffsets&&(e.styles.popper=Object.assign({},e.styles.popper,Mt(Object.assign({},c,{offsets:e.modifiersData.popperOffsets,position:e.options.strategy,adaptive:r,roundOffsets:l})))),null!=e.modifiersData.arrow&&(e.styles.arrow=Object.assign({},e.styles.arrow,Mt(Object.assign({},c,{offsets:e.modifiersData.arrow,position:"absolute",adaptive:!1,roundOffsets:l})))),e.attributes.popper=Object.assign({},e.attributes.popper,{"data-popper-placement":e.placement})},data:{}},Ht={passive:!0},Rt={name:"eventListeners",enabled:!0,phase:"write",fn:function(){},effect:function(t){var e=t.state,i=t.instance,n=t.options,s=n.scroll,o=void 0===s||s,r=n.resize,a=void 0===r||r,l=dt(e.elements.popper),c=[].concat(e.scrollParents.reference,e.scrollParents.popper);return o&&c.forEach((function(t){t.addEventListener("scroll",i.update,Ht)})),a&&l.addEventListener("resize",i.update,Ht),function(){o&&c.forEach((function(t){t.removeEventListener("scroll",i.update,Ht)})),a&&l.removeEventListener("resize",i.update,Ht)}},data:{}},Bt={left:"right",right:"left",bottom:"top",top:"bottom"};function Wt(t){return t.replace(/left|right|bottom|top/g,(function(t){return Bt[t]}))}var qt={start:"end",end:"start"};function zt(t){return t.replace(/start|end/g,(function(t){return qt[t]}))}function $t(t){var e=dt(t);return{scrollLeft:e.pageXOffset,scrollTop:e.pageYOffset}}function Ut(t){return _t(Et(t)).left+$t(t).scrollLeft}function Ft(t){var e=yt(t),i=e.overflow,n=e.overflowX,s=e.overflowY;return/auto|scroll|overlay|hidden/.test(i+s+n)}function Vt(t,e){var i;void 0===e&&(e=[]);var n=function t(e){return["html","body","#document"].indexOf(ht(e))>=0?e.ownerDocument.body:ft(e)&&Ft(e)?e:t(At(e))}(t),s=n===(null==(i=t.ownerDocument)?void 0:i.body),o=dt(n),r=s?[o].concat(o.visualViewport||[],Ft(n)?n:[]):n,a=e.concat(r);return s?a:a.concat(Vt(At(r)))}function Kt(t){return Object.assign({},t,{left:t.x,top:t.y,right:t.x+t.width,bottom:t.y+t.height})}function Xt(t,e){return"viewport"===e?Kt(function(t){var e=dt(t),i=Et(t),n=e.visualViewport,s=i.clientWidth,o=i.clientHeight,r=0,a=0;return n&&(s=n.width,o=n.height,/^((?!chrome|android).)*safari/i.test(navigator.userAgent)||(r=n.offsetLeft,a=n.offsetTop)),{width:s,height:o,x:r+Ut(t),y:a}}(t)):ft(e)?function(t){var e=_t(t);return e.top=e.top+t.clientTop,e.left=e.left+t.clientLeft,e.bottom=e.top+t.clientHeight,e.right=e.left+t.clientWidth,e.width=t.clientWidth,e.height=t.clientHeight,e.x=e.left,e.y=e.top,e}(e):Kt(function(t){var e,i=Et(t),n=$t(t),s=null==(e=t.ownerDocument)?void 0:e.body,o=kt(i.scrollWidth,i.clientWidth,s?s.scrollWidth:0,s?s.clientWidth:0),r=kt(i.scrollHeight,i.clientHeight,s?s.scrollHeight:0,s?s.clientHeight:0),a=-n.scrollLeft+Ut(t),l=-n.scrollTop;return"rtl"===yt(s||i).direction&&(a+=kt(i.clientWidth,s?s.clientWidth:0)-o),{width:o,height:r,x:a,y:l}}(Et(t)))}function Yt(t){return t.split("-")[1]}function Qt(t){var e,i=t.reference,n=t.element,s=t.placement,o=s?gt(s):null,r=s?Yt(s):null,a=i.x+i.width/2-n.width/2,l=i.y+i.height/2-n.height/2;switch(o){case it:e={x:a,y:i.y-n.height};break;case nt:e={x:a,y:i.y+i.height};break;case st:e={x:i.x+i.width,y:l};break;case ot:e={x:i.x-n.width,y:l};break;default:e={x:i.x,y:i.y}}var c=o?Ct(o):null;if(null!=c){var h="y"===c?"height":"width";switch(r){case"start":e[c]=e[c]-(i[h]/2-n[h]/2);break;case"end":e[c]=e[c]+(i[h]/2-n[h]/2)}}return e}function Gt(t,e){void 0===e&&(e={});var i=e,n=i.placement,s=void 0===n?t.placement:n,o=i.boundary,r=void 0===o?"clippingParents":o,a=i.rootBoundary,l=void 0===a?"viewport":a,c=i.elementContext,h=void 0===c?"popper":c,d=i.altBoundary,u=void 0!==d&&d,f=i.padding,p=void 0===f?0:f,m=St("number"!=typeof p?p:It(p,rt)),g="popper"===h?"reference":"popper",_=t.elements.reference,b=t.rects.popper,v=t.elements[u?g:h],y=function(t,e,i){var n="clippingParents"===e?function(t){var e=Vt(At(t)),i=["absolute","fixed"].indexOf(yt(t).position)>=0&&ft(t)?Ot(t):t;return ut(i)?e.filter((function(t){return ut(t)&&vt(t,i)&&"body"!==ht(t)})):[]}(t):[].concat(e),s=[].concat(n,[i]),o=s[0],r=s.reduce((function(e,i){var n=Xt(t,i);return e.top=kt(n.top,e.top),e.right=Lt(n.right,e.right),e.bottom=Lt(n.bottom,e.bottom),e.left=kt(n.left,e.left),e}),Xt(t,o));return r.width=r.right-r.left,r.height=r.bottom-r.top,r.x=r.left,r.y=r.top,r}(ut(v)?v:v.contextElement||Et(t.elements.popper),r,l),w=_t(_),E=Qt({reference:w,element:b,strategy:"absolute",placement:s}),A=Kt(Object.assign({},b,E)),T="popper"===h?A:w,O={top:y.top-T.top+m.top,bottom:T.bottom-y.bottom+m.bottom,left:y.left-T.left+m.left,right:T.right-y.right+m.right},C=t.modifiersData.offset;if("popper"===h&&C){var k=C[s];Object.keys(O).forEach((function(t){var e=[st,nt].indexOf(t)>=0?1:-1,i=[it,nt].indexOf(t)>=0?"y":"x";O[t]+=k[i]*e}))}return O}function Zt(t,e){void 0===e&&(e={});var i=e,n=i.placement,s=i.boundary,o=i.rootBoundary,r=i.padding,a=i.flipVariations,l=i.allowedAutoPlacements,c=void 0===l?lt:l,h=Yt(n),d=h?a?at:at.filter((function(t){return Yt(t)===h})):rt,u=d.filter((function(t){return c.indexOf(t)>=0}));0===u.length&&(u=d);var f=u.reduce((function(e,i){return e[i]=Gt(t,{placement:i,boundary:s,rootBoundary:o,padding:r})[gt(i)],e}),{});return Object.keys(f).sort((function(t,e){return f[t]-f[e]}))}var Jt={name:"flip",enabled:!0,phase:"main",fn:function(t){var e=t.state,i=t.options,n=t.name;if(!e.modifiersData[n]._skip){for(var s=i.mainAxis,o=void 0===s||s,r=i.altAxis,a=void 0===r||r,l=i.fallbackPlacements,c=i.padding,h=i.boundary,d=i.rootBoundary,u=i.altBoundary,f=i.flipVariations,p=void 0===f||f,m=i.allowedAutoPlacements,g=e.options.placement,_=gt(g),b=l||(_!==g&&p?function(t){if("auto"===gt(t))return[];var e=Wt(t);return[zt(t),e,zt(e)]}(g):[Wt(g)]),v=[g].concat(b).reduce((function(t,i){return t.concat("auto"===gt(i)?Zt(e,{placement:i,boundary:h,rootBoundary:d,padding:c,flipVariations:p,allowedAutoPlacements:m}):i)}),[]),y=e.rects.reference,w=e.rects.popper,E=new Map,A=!0,T=v[0],O=0;O<v.length;O++){var C=v[O],k=gt(C),L="start"===Yt(C),x=[it,nt].indexOf(k)>=0,D=x?"width":"height",S=Gt(e,{placement:C,boundary:h,rootBoundary:d,altBoundary:u,padding:c}),I=x?L?st:ot:L?nt:it;y[D]>w[D]&&(I=Wt(I));var N=Wt(I),j=[];if(o&&j.push(S[k]<=0),a&&j.push(S[I]<=0,S[N]<=0),j.every((function(t){return t}))){T=C,A=!1;break}E.set(C,j)}if(A)for(var M=function(t){var e=v.find((function(e){var i=E.get(e);if(i)return i.slice(0,t).every((function(t){return t}))}));if(e)return T=e,"break"},P=p?3:1;P>0&&"break"!==M(P);P--);e.placement!==T&&(e.modifiersData[n]._skip=!0,e.placement=T,e.reset=!0)}},requiresIfExists:["offset"],data:{_skip:!1}};function te(t,e,i){return void 0===i&&(i={x:0,y:0}),{top:t.top-e.height-i.y,right:t.right-e.width+i.x,bottom:t.bottom-e.height+i.y,left:t.left-e.width-i.x}}function ee(t){return[it,st,nt,ot].some((function(e){return t[e]>=0}))}var ie={name:"hide",enabled:!0,phase:"main",requiresIfExists:["preventOverflow"],fn:function(t){var e=t.state,i=t.name,n=e.rects.reference,s=e.rects.popper,o=e.modifiersData.preventOverflow,r=Gt(e,{elementContext:"reference"}),a=Gt(e,{altBoundary:!0}),l=te(r,n),c=te(a,s,o),h=ee(l),d=ee(c);e.modifiersData[i]={referenceClippingOffsets:l,popperEscapeOffsets:c,isReferenceHidden:h,hasPopperEscaped:d},e.attributes.popper=Object.assign({},e.attributes.popper,{"data-popper-reference-hidden":h,"data-popper-escaped":d})}},ne={name:"offset",enabled:!0,phase:"main",requires:["popperOffsets"],fn:function(t){var e=t.state,i=t.options,n=t.name,s=i.offset,o=void 0===s?[0,0]:s,r=lt.reduce((function(t,i){return t[i]=function(t,e,i){var n=gt(t),s=[ot,it].indexOf(n)>=0?-1:1,o="function"==typeof i?i(Object.assign({},e,{placement:t})):i,r=o[0],a=o[1];return r=r||0,a=(a||0)*s,[ot,st].indexOf(n)>=0?{x:a,y:r}:{x:r,y:a}}(i,e.rects,o),t}),{}),a=r[e.placement],l=a.x,c=a.y;null!=e.modifiersData.popperOffsets&&(e.modifiersData.popperOffsets.x+=l,e.modifiersData.popperOffsets.y+=c),e.modifiersData[n]=r}},se={name:"popperOffsets",enabled:!0,phase:"read",fn:function(t){var e=t.state,i=t.name;e.modifiersData[i]=Qt({reference:e.rects.reference,element:e.rects.popper,strategy:"absolute",placement:e.placement})},data:{}},oe={name:"preventOverflow",enabled:!0,phase:"main",fn:function(t){var e=t.state,i=t.options,n=t.name,s=i.mainAxis,o=void 0===s||s,r=i.altAxis,a=void 0!==r&&r,l=i.boundary,c=i.rootBoundary,h=i.altBoundary,d=i.padding,u=i.tether,f=void 0===u||u,p=i.tetherOffset,m=void 0===p?0:p,g=Gt(e,{boundary:l,rootBoundary:c,padding:d,altBoundary:h}),_=gt(e.placement),b=Yt(e.placement),v=!b,y=Ct(_),w="x"===y?"y":"x",E=e.modifiersData.popperOffsets,A=e.rects.reference,T=e.rects.popper,O="function"==typeof m?m(Object.assign({},e.rects,{placement:e.placement})):m,C={x:0,y:0};if(E){if(o||a){var k="y"===y?it:ot,L="y"===y?nt:st,x="y"===y?"height":"width",D=E[y],S=E[y]+g[k],I=E[y]-g[L],N=f?-T[x]/2:0,j="start"===b?A[x]:T[x],M="start"===b?-T[x]:-A[x],P=e.elements.arrow,H=f&&P?bt(P):{width:0,height:0},R=e.modifiersData["arrow#persistent"]?e.modifiersData["arrow#persistent"].padding:{top:0,right:0,bottom:0,left:0},B=R[k],W=R[L],q=Dt(0,A[x],H[x]),z=v?A[x]/2-N-q-B-O:j-q-B-O,$=v?-A[x]/2+N+q+W+O:M+q+W+O,U=e.elements.arrow&&Ot(e.elements.arrow),F=U?"y"===y?U.clientTop||0:U.clientLeft||0:0,V=e.modifiersData.offset?e.modifiersData.offset[e.placement][y]:0,K=E[y]+z-V-F,X=E[y]+$-V;if(o){var Y=Dt(f?Lt(S,K):S,D,f?kt(I,X):I);E[y]=Y,C[y]=Y-D}if(a){var Q="x"===y?it:ot,G="x"===y?nt:st,Z=E[w],J=Z+g[Q],tt=Z-g[G],et=Dt(f?Lt(J,K):J,Z,f?kt(tt,X):tt);E[w]=et,C[w]=et-Z}}e.modifiersData[n]=C}},requiresIfExists:["offset"]};function re(t,e,i){void 0===i&&(i=!1);var n,s,o=Et(e),r=_t(t),a=ft(e),l={scrollLeft:0,scrollTop:0},c={x:0,y:0};return(a||!a&&!i)&&(("body"!==ht(e)||Ft(o))&&(l=(n=e)!==dt(n)&&ft(n)?{scrollLeft:(s=n).scrollLeft,scrollTop:s.scrollTop}:$t(n)),ft(e)?((c=_t(e)).x+=e.clientLeft,c.y+=e.clientTop):o&&(c.x=Ut(o))),{x:r.left+l.scrollLeft-c.x,y:r.top+l.scrollTop-c.y,width:r.width,height:r.height}}var ae={placement:"bottom",modifiers:[],strategy:"absolute"};function le(){for(var t=arguments.length,e=new Array(t),i=0;i<t;i++)e[i]=arguments[i];return!e.some((function(t){return!(t&&"function"==typeof t.getBoundingClientRect)}))}function ce(t){void 0===t&&(t={});var e=t,i=e.defaultModifiers,n=void 0===i?[]:i,s=e.defaultOptions,o=void 0===s?ae:s;return function(t,e,i){void 0===i&&(i=o);var s,r,a={placement:"bottom",orderedModifiers:[],options:Object.assign({},ae,o),modifiersData:{},elements:{reference:t,popper:e},attributes:{},styles:{}},l=[],c=!1,h={state:a,setOptions:function(i){d(),a.options=Object.assign({},o,a.options,i),a.scrollParents={reference:ut(t)?Vt(t):t.contextElement?Vt(t.contextElement):[],popper:Vt(e)};var s,r,c=function(t){var e=function(t){var e=new Map,i=new Set,n=[];return t.forEach((function(t){e.set(t.name,t)})),t.forEach((function(t){i.has(t.name)||function t(s){i.add(s.name),[].concat(s.requires||[],s.requiresIfExists||[]).forEach((function(n){if(!i.has(n)){var s=e.get(n);s&&t(s)}})),n.push(s)}(t)})),n}(t);return ct.reduce((function(t,i){return t.concat(e.filter((function(t){return t.phase===i})))}),[])}((s=[].concat(n,a.options.modifiers),r=s.reduce((function(t,e){var i=t[e.name];return t[e.name]=i?Object.assign({},i,e,{options:Object.assign({},i.options,e.options),data:Object.assign({},i.data,e.data)}):e,t}),{}),Object.keys(r).map((function(t){return r[t]}))));return a.orderedModifiers=c.filter((function(t){return t.enabled})),a.orderedModifiers.forEach((function(t){var e=t.name,i=t.options,n=void 0===i?{}:i,s=t.effect;if("function"==typeof s){var o=s({state:a,name:e,instance:h,options:n});l.push(o||function(){})}})),h.update()},forceUpdate:function(){if(!c){var t=a.elements,e=t.reference,i=t.popper;if(le(e,i)){a.rects={reference:re(e,Ot(i),"fixed"===a.options.strategy),popper:bt(i)},a.reset=!1,a.placement=a.options.placement,a.orderedModifiers.forEach((function(t){return a.modifiersData[t.name]=Object.assign({},t.data)}));for(var n=0;n<a.orderedModifiers.length;n++)if(!0!==a.reset){var s=a.orderedModifiers[n],o=s.fn,r=s.options,l=void 0===r?{}:r,d=s.name;"function"==typeof o&&(a=o({state:a,options:l,name:d,instance:h})||a)}else a.reset=!1,n=-1}}},update:(s=function(){return new Promise((function(t){h.forceUpdate(),t(a)}))},function(){return r||(r=new Promise((function(t){Promise.resolve().then((function(){r=void 0,t(s())}))}))),r}),destroy:function(){d(),c=!0}};if(!le(t,e))return h;function d(){l.forEach((function(t){return t()})),l=[]}return h.setOptions(i).then((function(t){!c&&i.onFirstUpdate&&i.onFirstUpdate(t)})),h}}var he=ce(),de=ce({defaultModifiers:[Rt,se,Pt,mt]}),ue=ce({defaultModifiers:[Rt,se,Pt,mt,ne,Jt,oe,Nt,ie]}),fe=Object.freeze({__proto__:null,popperGenerator:ce,detectOverflow:Gt,createPopperBase:he,createPopper:ue,createPopperLite:de,top:it,bottom:nt,right:st,left:ot,auto:"auto",basePlacements:rt,start:"start",end:"end",clippingParents:"clippingParents",viewport:"viewport",popper:"popper",reference:"reference",variationPlacements:at,placements:lt,beforeRead:"beforeRead",read:"read",afterRead:"afterRead",beforeMain:"beforeMain",main:"main",afterMain:"afterMain",beforeWrite:"beforeWrite",write:"write",afterWrite:"afterWrite",modifierPhases:ct,applyStyles:mt,arrow:Nt,computeStyles:Pt,eventListeners:Rt,flip:Jt,hide:ie,offset:ne,popperOffsets:se,preventOverflow:oe});const pe=new RegExp("ArrowUp|ArrowDown|Escape"),me=g()?"top-end":"top-start",ge=g()?"top-start":"top-end",_e=g()?"bottom-end":"bottom-start",be=g()?"bottom-start":"bottom-end",ve=g()?"left-start":"right-start",ye=g()?"right-start":"left-start",we={offset:[0,2],boundary:"clippingParents",reference:"toggle",display:"dynamic",popperConfig:null,autoClose:!0},Ee={offset:"(array|string|function)",boundary:"(string|element)",reference:"(string|element|object)",display:"string",popperConfig:"(null|object|function)",autoClose:"(boolean|string)"};class Ae extends B{constructor(t,e){super(t),this._popper=null,this._config=this._getConfig(e),this._menu=this._getMenuElement(),this._inNavbar=this._detectNavbar(),this._addEventListeners()}static get Default(){return we}static get DefaultType(){return Ee}static get NAME(){return"dropdown"}toggle(){h(this._element)||(this._element.classList.contains("show")?this.hide():this.show())}show(){if(h(this._element)||this._menu.classList.contains("show"))return;const t=Ae.getParentFromElement(this._element),e={relatedTarget:this._element};if(!P.trigger(this._element,"show.bs.dropdown",e).defaultPrevented){if(this._inNavbar)U.setDataAttribute(this._menu,"popper","none");else{if(void 0===fe)throw new TypeError("Bootstrap's dropdowns require Popper (https://popper.js.org)");let e=this._element;"parent"===this._config.reference?e=t:r(this._config.reference)?e=a(this._config.reference):"object"==typeof this._config.reference&&(e=this._config.reference);const i=this._getPopperConfig(),n=i.modifiers.find(t=>"applyStyles"===t.name&&!1===t.enabled);this._popper=ue(e,this._menu,i),n&&U.setDataAttribute(this._menu,"popper","static")}"ontouchstart"in document.documentElement&&!t.closest(".navbar-nav")&&[].concat(...document.body.children).forEach(t=>P.on(t,"mouseover",u)),this._element.focus(),this._element.setAttribute("aria-expanded",!0),this._menu.classList.toggle("show"),this._element.classList.toggle("show"),P.trigger(this._element,"shown.bs.dropdown",e)}}hide(){if(h(this._element)||!this._menu.classList.contains("show"))return;const t={relatedTarget:this._element};this._completeHide(t)}dispose(){this._popper&&this._popper.destroy(),super.dispose()}update(){this._inNavbar=this._detectNavbar(),this._popper&&this._popper.update()}_addEventListeners(){P.on(this._element,"click.bs.dropdown",t=>{t.preventDefault(),this.toggle()})}_completeHide(t){P.trigger(this._element,"hide.bs.dropdown",t).defaultPrevented||("ontouchstart"in document.documentElement&&[].concat(...document.body.children).forEach(t=>P.off(t,"mouseover",u)),this._popper&&this._popper.destroy(),this._menu.classList.remove("show"),this._element.classList.remove("show"),this._element.setAttribute("aria-expanded","false"),U.removeDataAttribute(this._menu,"popper"),P.trigger(this._element,"hidden.bs.dropdown",t))}_getConfig(t){if(t={...this.constructor.Default,...U.getDataAttributes(this._element),...t},l("dropdown",t,this.constructor.DefaultType),"object"==typeof t.reference&&!r(t.reference)&&"function"!=typeof t.reference.getBoundingClientRect)throw new TypeError("dropdown".toUpperCase()+': Option "reference" provided type "object" without a required "getBoundingClientRect" method.');return t}_getMenuElement(){return t.next(this._element,".dropdown-menu")[0]}_getPlacement(){const t=this._element.parentNode;if(t.classList.contains("dropend"))return ve;if(t.classList.contains("dropstart"))return ye;const e="end"===getComputedStyle(this._menu).getPropertyValue("--bs-position").trim();return t.classList.contains("dropup")?e?ge:me:e?be:_e}_detectNavbar(){return null!==this._element.closest(".navbar")}_getOffset(){const{offset:t}=this._config;return"string"==typeof t?t.split(",").map(t=>Number.parseInt(t,10)):"function"==typeof t?e=>t(e,this._element):t}_getPopperConfig(){const t={placement:this._getPlacement(),modifiers:[{name:"preventOverflow",options:{boundary:this._config.boundary}},{name:"offset",options:{offset:this._getOffset()}}]};return"static"===this._config.display&&(t.modifiers=[{name:"applyStyles",enabled:!1}]),{...t,..."function"==typeof this._config.popperConfig?this._config.popperConfig(t):this._config.popperConfig}}_selectMenuItem({key:e,target:i}){const n=t.find(".dropdown-menu .dropdown-item:not(.disabled):not(:disabled)",this._menu).filter(c);n.length&&y(n,i,"ArrowDown"===e,!n.includes(i)).focus()}static dropdownInterface(t,e){const i=Ae.getOrCreateInstance(t,e);if("string"==typeof e){if(void 0===i[e])throw new TypeError(`No method named "${e}"`);i[e]()}}static jQueryInterface(t){return this.each((function(){Ae.dropdownInterface(this,t)}))}static clearMenus(e){if(e&&(2===e.button||"keyup"===e.type&&"Tab"!==e.key))return;const i=t.find('[data-bs-toggle="dropdown"]');for(let t=0,n=i.length;t<n;t++){const n=Ae.getInstance(i[t]);if(!n||!1===n._config.autoClose)continue;if(!n._element.classList.contains("show"))continue;const s={relatedTarget:n._element};if(e){const t=e.composedPath(),i=t.includes(n._menu);if(t.includes(n._element)||"inside"===n._config.autoClose&&!i||"outside"===n._config.autoClose&&i)continue;if(n._menu.contains(e.target)&&("keyup"===e.type&&"Tab"===e.key||/input|select|option|textarea|form/i.test(e.target.tagName)))continue;"click"===e.type&&(s.clickEvent=e)}n._completeHide(s)}}static getParentFromElement(t){return s(t)||t.parentNode}static dataApiKeydownHandler(e){if(/input|textarea/i.test(e.target.tagName)?"Space"===e.key||"Escape"!==e.key&&("ArrowDown"!==e.key&&"ArrowUp"!==e.key||e.target.closest(".dropdown-menu")):!pe.test(e.key))return;const i=this.classList.contains("show");if(!i&&"Escape"===e.key)return;if(e.preventDefault(),e.stopPropagation(),h(this))return;const n=()=>this.matches('[data-bs-toggle="dropdown"]')?this:t.prev(this,'[data-bs-toggle="dropdown"]')[0];return"Escape"===e.key?(n().focus(),void Ae.clearMenus()):"ArrowUp"===e.key||"ArrowDown"===e.key?(i||n().click(),void Ae.getInstance(n())._selectMenuItem(e)):void(i&&"Space"!==e.key||Ae.clearMenus())}}P.on(document,"keydown.bs.dropdown.data-api",'[data-bs-toggle="dropdown"]',Ae.dataApiKeydownHandler),P.on(document,"keydown.bs.dropdown.data-api",".dropdown-menu",Ae.dataApiKeydownHandler),P.on(document,"click.bs.dropdown.data-api",Ae.clearMenus),P.on(document,"keyup.bs.dropdown.data-api",Ae.clearMenus),P.on(document,"click.bs.dropdown.data-api",'[data-bs-toggle="dropdown"]',(function(t){t.preventDefault(),Ae.dropdownInterface(this)})),_(Ae);class Te{constructor(){this._element=document.body}getWidth(){const t=document.documentElement.clientWidth;return Math.abs(window.innerWidth-t)}hide(){const t=this.getWidth();this._disableOverFlow(),this._setElementAttributes(this._element,"paddingRight",e=>e+t),this._setElementAttributes(".fixed-top, .fixed-bottom, .is-fixed, .sticky-top","paddingRight",e=>e+t),this._setElementAttributes(".sticky-top","marginRight",e=>e-t)}_disableOverFlow(){this._saveInitialAttribute(this._element,"overflow"),this._element.style.overflow="hidden"}_setElementAttributes(t,e,i){const n=this.getWidth();this._applyManipulationCallback(t,t=>{if(t!==this._element&&window.innerWidth>t.clientWidth+n)return;this._saveInitialAttribute(t,e);const s=window.getComputedStyle(t)[e];t.style[e]=i(Number.parseFloat(s))+"px"})}reset(){this._resetElementAttributes(this._element,"overflow"),this._resetElementAttributes(this._element,"paddingRight"),this._resetElementAttributes(".fixed-top, .fixed-bottom, .is-fixed, .sticky-top","paddingRight"),this._resetElementAttributes(".sticky-top","marginRight")}_saveInitialAttribute(t,e){const i=t.style[e];i&&U.setDataAttribute(t,e,i)}_resetElementAttributes(t,e){this._applyManipulationCallback(t,t=>{const i=U.getDataAttribute(t,e);void 0===i?t.style.removeProperty(e):(U.removeDataAttribute(t,e),t.style[e]=i)})}_applyManipulationCallback(e,i){r(e)?i(e):t.find(e,this._element).forEach(i)}isOverflowing(){return this.getWidth()>0}}const Oe={isVisible:!0,isAnimated:!1,rootElement:"body",clickCallback:null},Ce={isVisible:"boolean",isAnimated:"boolean",rootElement:"(element|string)",clickCallback:"(function|null)"};class ke{constructor(t){this._config=this._getConfig(t),this._isAppended=!1,this._element=null}show(t){this._config.isVisible?(this._append(),this._config.isAnimated&&f(this._getElement()),this._getElement().classList.add("show"),this._emulateAnimation(()=>{b(t)})):b(t)}hide(t){this._config.isVisible?(this._getElement().classList.remove("show"),this._emulateAnimation(()=>{this.dispose(),b(t)})):b(t)}_getElement(){if(!this._element){const t=document.createElement("div");t.className="modal-backdrop",this._config.isAnimated&&t.classList.add("fade"),this._element=t}return this._element}_getConfig(t){return(t={...Oe,..."object"==typeof t?t:{}}).rootElement=a(t.rootElement),l("backdrop",t,Ce),t}_append(){this._isAppended||(this._config.rootElement.appendChild(this._getElement()),P.on(this._getElement(),"mousedown.bs.backdrop",()=>{b(this._config.clickCallback)}),this._isAppended=!0)}dispose(){this._isAppended&&(P.off(this._element,"mousedown.bs.backdrop"),this._element.remove(),this._isAppended=!1)}_emulateAnimation(t){v(t,this._getElement(),this._config.isAnimated)}}const Le={backdrop:!0,keyboard:!0,focus:!0},xe={backdrop:"(boolean|string)",keyboard:"boolean",focus:"boolean"};class De extends B{constructor(e,i){super(e),this._config=this._getConfig(i),this._dialog=t.findOne(".modal-dialog",this._element),this._backdrop=this._initializeBackDrop(),this._isShown=!1,this._ignoreBackdropClick=!1,this._isTransitioning=!1,this._scrollBar=new Te}static get Default(){return Le}static get NAME(){return"modal"}toggle(t){return this._isShown?this.hide():this.show(t)}show(t){this._isShown||this._isTransitioning||P.trigger(this._element,"show.bs.modal",{relatedTarget:t}).defaultPrevented||(this._isShown=!0,this._isAnimated()&&(this._isTransitioning=!0),this._scrollBar.hide(),document.body.classList.add("modal-open"),this._adjustDialog(),this._setEscapeEvent(),this._setResizeEvent(),P.on(this._element,"click.dismiss.bs.modal",'[data-bs-dismiss="modal"]',t=>this.hide(t)),P.on(this._dialog,"mousedown.dismiss.bs.modal",()=>{P.one(this._element,"mouseup.dismiss.bs.modal",t=>{t.target===this._element&&(this._ignoreBackdropClick=!0)})}),this._showBackdrop(()=>this._showElement(t)))}hide(t){if(t&&["A","AREA"].includes(t.target.tagName)&&t.preventDefault(),!this._isShown||this._isTransitioning)return;if(P.trigger(this._element,"hide.bs.modal").defaultPrevented)return;this._isShown=!1;const e=this._isAnimated();e&&(this._isTransitioning=!0),this._setEscapeEvent(),this._setResizeEvent(),P.off(document,"focusin.bs.modal"),this._element.classList.remove("show"),P.off(this._element,"click.dismiss.bs.modal"),P.off(this._dialog,"mousedown.dismiss.bs.modal"),this._queueCallback(()=>this._hideModal(),this._element,e)}dispose(){[window,this._dialog].forEach(t=>P.off(t,".bs.modal")),this._backdrop.dispose(),super.dispose(),P.off(document,"focusin.bs.modal")}handleUpdate(){this._adjustDialog()}_initializeBackDrop(){return new ke({isVisible:Boolean(this._config.backdrop),isAnimated:this._isAnimated()})}_getConfig(t){return t={...Le,...U.getDataAttributes(this._element),..."object"==typeof t?t:{}},l("modal",t,xe),t}_showElement(e){const i=this._isAnimated(),n=t.findOne(".modal-body",this._dialog);this._element.parentNode&&this._element.parentNode.nodeType===Node.ELEMENT_NODE||document.body.appendChild(this._element),this._element.style.display="block",this._element.removeAttribute("aria-hidden"),this._element.setAttribute("aria-modal",!0),this._element.setAttribute("role","dialog"),this._element.scrollTop=0,n&&(n.scrollTop=0),i&&f(this._element),this._element.classList.add("show"),this._config.focus&&this._enforceFocus(),this._queueCallback(()=>{this._config.focus&&this._element.focus(),this._isTransitioning=!1,P.trigger(this._element,"shown.bs.modal",{relatedTarget:e})},this._dialog,i)}_enforceFocus(){P.off(document,"focusin.bs.modal"),P.on(document,"focusin.bs.modal",t=>{document===t.target||this._element===t.target||this._element.contains(t.target)||this._element.focus()})}_setEscapeEvent(){this._isShown?P.on(this._element,"keydown.dismiss.bs.modal",t=>{this._config.keyboard&&"Escape"===t.key?(t.preventDefault(),this.hide()):this._config.keyboard||"Escape"!==t.key||this._triggerBackdropTransition()}):P.off(this._element,"keydown.dismiss.bs.modal")}_setResizeEvent(){this._isShown?P.on(window,"resize.bs.modal",()=>this._adjustDialog()):P.off(window,"resize.bs.modal")}_hideModal(){this._element.style.display="none",this._element.setAttribute("aria-hidden",!0),this._element.removeAttribute("aria-modal"),this._element.removeAttribute("role"),this._isTransitioning=!1,this._backdrop.hide(()=>{document.body.classList.remove("modal-open"),this._resetAdjustments(),this._scrollBar.reset(),P.trigger(this._element,"hidden.bs.modal")})}_showBackdrop(t){P.on(this._element,"click.dismiss.bs.modal",t=>{this._ignoreBackdropClick?this._ignoreBackdropClick=!1:t.target===t.currentTarget&&(!0===this._config.backdrop?this.hide():"static"===this._config.backdrop&&this._triggerBackdropTransition())}),this._backdrop.show(t)}_isAnimated(){return this._element.classList.contains("fade")}_triggerBackdropTransition(){if(P.trigger(this._element,"hidePrevented.bs.modal").defaultPrevented)return;const{classList:t,scrollHeight:e,style:i}=this._element,n=e>document.documentElement.clientHeight;!n&&"hidden"===i.overflowY||t.contains("modal-static")||(n||(i.overflowY="hidden"),t.add("modal-static"),this._queueCallback(()=>{t.remove("modal-static"),n||this._queueCallback(()=>{i.overflowY=""},this._dialog)},this._dialog),this._element.focus())}_adjustDialog(){const t=this._element.scrollHeight>document.documentElement.clientHeight,e=this._scrollBar.getWidth(),i=e>0;(!i&&t&&!g()||i&&!t&&g())&&(this._element.style.paddingLeft=e+"px"),(i&&!t&&!g()||!i&&t&&g())&&(this._element.style.paddingRight=e+"px")}_resetAdjustments(){this._element.style.paddingLeft="",this._element.style.paddingRight=""}static jQueryInterface(t,e){return this.each((function(){const i=De.getOrCreateInstance(this,t);if("string"==typeof t){if(void 0===i[t])throw new TypeError(`No method named "${t}"`);i[t](e)}}))}}P.on(document,"click.bs.modal.data-api",'[data-bs-toggle="modal"]',(function(t){const e=s(this);["A","AREA"].includes(this.tagName)&&t.preventDefault(),P.one(e,"show.bs.modal",t=>{t.defaultPrevented||P.one(e,"hidden.bs.modal",()=>{c(this)&&this.focus()})}),De.getOrCreateInstance(e).toggle(this)})),_(De);const Se={backdrop:!0,keyboard:!0,scroll:!1},Ie={backdrop:"boolean",keyboard:"boolean",scroll:"boolean"};class Ne extends B{constructor(t,e){super(t),this._config=this._getConfig(e),this._isShown=!1,this._backdrop=this._initializeBackDrop(),this._addEventListeners()}static get NAME(){return"offcanvas"}static get Default(){return Se}toggle(t){return this._isShown?this.hide():this.show(t)}show(t){this._isShown||P.trigger(this._element,"show.bs.offcanvas",{relatedTarget:t}).defaultPrevented||(this._isShown=!0,this._element.style.visibility="visible",this._backdrop.show(),this._config.scroll||((new Te).hide(),this._enforceFocusOnElement(this._element)),this._element.removeAttribute("aria-hidden"),this._element.setAttribute("aria-modal",!0),this._element.setAttribute("role","dialog"),this._element.classList.add("show"),this._queueCallback(()=>{P.trigger(this._element,"shown.bs.offcanvas",{relatedTarget:t})},this._element,!0))}hide(){this._isShown&&(P.trigger(this._element,"hide.bs.offcanvas").defaultPrevented||(P.off(document,"focusin.bs.offcanvas"),this._element.blur(),this._isShown=!1,this._element.classList.remove("show"),this._backdrop.hide(),this._queueCallback(()=>{this._element.setAttribute("aria-hidden",!0),this._element.removeAttribute("aria-modal"),this._element.removeAttribute("role"),this._element.style.visibility="hidden",this._config.scroll||(new Te).reset(),P.trigger(this._element,"hidden.bs.offcanvas")},this._element,!0)))}dispose(){this._backdrop.dispose(),super.dispose(),P.off(document,"focusin.bs.offcanvas")}_getConfig(t){return t={...Se,...U.getDataAttributes(this._element),..."object"==typeof t?t:{}},l("offcanvas",t,Ie),t}_initializeBackDrop(){return new ke({isVisible:this._config.backdrop,isAnimated:!0,rootElement:this._element.parentNode,clickCallback:()=>this.hide()})}_enforceFocusOnElement(t){P.off(document,"focusin.bs.offcanvas"),P.on(document,"focusin.bs.offcanvas",e=>{document===e.target||t===e.target||t.contains(e.target)||t.focus()}),t.focus()}_addEventListeners(){P.on(this._element,"click.dismiss.bs.offcanvas",'[data-bs-dismiss="offcanvas"]',()=>this.hide()),P.on(this._element,"keydown.dismiss.bs.offcanvas",t=>{this._config.keyboard&&"Escape"===t.key&&this.hide()})}static jQueryInterface(t){return this.each((function(){const e=Ne.getOrCreateInstance(this,t);if("string"==typeof t){if(void 0===e[t]||t.startsWith("_")||"constructor"===t)throw new TypeError(`No method named "${t}"`);e[t](this)}}))}}P.on(document,"click.bs.offcanvas.data-api",'[data-bs-toggle="offcanvas"]',(function(e){const i=s(this);if(["A","AREA"].includes(this.tagName)&&e.preventDefault(),h(this))return;P.one(i,"hidden.bs.offcanvas",()=>{c(this)&&this.focus()});const n=t.findOne(".offcanvas.show");n&&n!==i&&Ne.getInstance(n).hide(),Ne.getOrCreateInstance(i).toggle(this)})),P.on(window,"load.bs.offcanvas.data-api",()=>t.find(".offcanvas.show").forEach(t=>Ne.getOrCreateInstance(t).show())),_(Ne);const je=new Set(["background","cite","href","itemtype","longdesc","poster","src","xlink:href"]),Me=/^(?:(?:https?|mailto|ftp|tel|file):|[^#&/:?]*(?:[#/?]|$))/i,Pe=/^data:(?:image\/(?:bmp|gif|jpeg|jpg|png|tiff|webp)|video\/(?:mpeg|mp4|ogg|webm)|audio\/(?:mp3|oga|ogg|opus));base64,[\d+/a-z]+=*$/i,He=(t,e)=>{const i=t.nodeName.toLowerCase();if(e.includes(i))return!je.has(i)||Boolean(Me.test(t.nodeValue)||Pe.test(t.nodeValue));const n=e.filter(t=>t instanceof RegExp);for(let t=0,e=n.length;t<e;t++)if(n[t].test(i))return!0;return!1};function Re(t,e,i){if(!t.length)return t;if(i&&"function"==typeof i)return i(t);const n=(new window.DOMParser).parseFromString(t,"text/html"),s=Object.keys(e),o=[].concat(...n.body.querySelectorAll("*"));for(let t=0,i=o.length;t<i;t++){const i=o[t],n=i.nodeName.toLowerCase();if(!s.includes(n)){i.remove();continue}const r=[].concat(...i.attributes),a=[].concat(e["*"]||[],e[n]||[]);r.forEach(t=>{He(t,a)||i.removeAttribute(t.nodeName)})}return n.body.innerHTML}const Be=new RegExp("(^|\\s)bs-tooltip\\S+","g"),We=new Set(["sanitize","allowList","sanitizeFn"]),qe={animation:"boolean",template:"string",title:"(string|element|function)",trigger:"string",delay:"(number|object)",html:"boolean",selector:"(string|boolean)",placement:"(string|function)",offset:"(array|string|function)",container:"(string|element|boolean)",fallbackPlacements:"array",boundary:"(string|element)",customClass:"(string|function)",sanitize:"boolean",sanitizeFn:"(null|function)",allowList:"object",popperConfig:"(null|object|function)"},ze={AUTO:"auto",TOP:"top",RIGHT:g()?"left":"right",BOTTOM:"bottom",LEFT:g()?"right":"left"},$e={animation:!0,template:'<div class="tooltip" role="tooltip"><div class="tooltip-arrow"></div><div class="tooltip-inner"></div></div>',trigger:"hover focus",title:"",delay:0,html:!1,selector:!1,placement:"top",offset:[0,0],container:!1,fallbackPlacements:["top","right","bottom","left"],boundary:"clippingParents",customClass:"",sanitize:!0,sanitizeFn:null,allowList:{"*":["class","dir","id","lang","role",/^aria-[\w-]*$/i],a:["target","href","title","rel"],area:[],b:[],br:[],col:[],code:[],div:[],em:[],hr:[],h1:[],h2:[],h3:[],h4:[],h5:[],h6:[],i:[],img:["src","srcset","alt","title","width","height"],li:[],ol:[],p:[],pre:[],s:[],small:[],span:[],sub:[],sup:[],strong:[],u:[],ul:[]},popperConfig:null},Ue={HIDE:"hide.bs.tooltip",HIDDEN:"hidden.bs.tooltip",SHOW:"show.bs.tooltip",SHOWN:"shown.bs.tooltip",INSERTED:"inserted.bs.tooltip",CLICK:"click.bs.tooltip",FOCUSIN:"focusin.bs.tooltip",FOCUSOUT:"focusout.bs.tooltip",MOUSEENTER:"mouseenter.bs.tooltip",MOUSELEAVE:"mouseleave.bs.tooltip"};class Fe extends B{constructor(t,e){if(void 0===fe)throw new TypeError("Bootstrap's tooltips require Popper (https://popper.js.org)");super(t),this._isEnabled=!0,this._timeout=0,this._hoverState="",this._activeTrigger={},this._popper=null,this._config=this._getConfig(e),this.tip=null,this._setListeners()}static get Default(){return $e}static get NAME(){return"tooltip"}static get Event(){return Ue}static get DefaultType(){return qe}enable(){this._isEnabled=!0}disable(){this._isEnabled=!1}toggleEnabled(){this._isEnabled=!this._isEnabled}toggle(t){if(this._isEnabled)if(t){const e=this._initializeOnDelegatedTarget(t);e._activeTrigger.click=!e._activeTrigger.click,e._isWithActiveTrigger()?e._enter(null,e):e._leave(null,e)}else{if(this.getTipElement().classList.contains("show"))return void this._leave(null,this);this._enter(null,this)}}dispose(){clearTimeout(this._timeout),P.off(this._element.closest(".modal"),"hide.bs.modal",this._hideModalHandler),this.tip&&this.tip.remove(),this._popper&&this._popper.destroy(),super.dispose()}show(){if("none"===this._element.style.display)throw new Error("Please use show on visible elements");if(!this.isWithContent()||!this._isEnabled)return;const t=P.trigger(this._element,this.constructor.Event.SHOW),i=d(this._element),n=null===i?this._element.ownerDocument.documentElement.contains(this._element):i.contains(this._element);if(t.defaultPrevented||!n)return;const s=this.getTipElement(),o=e(this.constructor.NAME);s.setAttribute("id",o),this._element.setAttribute("aria-describedby",o),this.setContent(),this._config.animation&&s.classList.add("fade");const r="function"==typeof this._config.placement?this._config.placement.call(this,s,this._element):this._config.placement,a=this._getAttachment(r);this._addAttachmentClass(a);const{container:l}=this._config;R.set(s,this.constructor.DATA_KEY,this),this._element.ownerDocument.documentElement.contains(this.tip)||(l.appendChild(s),P.trigger(this._element,this.constructor.Event.INSERTED)),this._popper?this._popper.update():this._popper=ue(this._element,s,this._getPopperConfig(a)),s.classList.add("show");const c="function"==typeof this._config.customClass?this._config.customClass():this._config.customClass;c&&s.classList.add(...c.split(" ")),"ontouchstart"in document.documentElement&&[].concat(...document.body.children).forEach(t=>{P.on(t,"mouseover",u)});const h=this.tip.classList.contains("fade");this._queueCallback(()=>{const t=this._hoverState;this._hoverState=null,P.trigger(this._element,this.constructor.Event.SHOWN),"out"===t&&this._leave(null,this)},this.tip,h)}hide(){if(!this._popper)return;const t=this.getTipElement();if(P.trigger(this._element,this.constructor.Event.HIDE).defaultPrevented)return;t.classList.remove("show"),"ontouchstart"in document.documentElement&&[].concat(...document.body.children).forEach(t=>P.off(t,"mouseover",u)),this._activeTrigger.click=!1,this._activeTrigger.focus=!1,this._activeTrigger.hover=!1;const e=this.tip.classList.contains("fade");this._queueCallback(()=>{this._isWithActiveTrigger()||("show"!==this._hoverState&&t.remove(),this._cleanTipClass(),this._element.removeAttribute("aria-describedby"),P.trigger(this._element,this.constructor.Event.HIDDEN),this._popper&&(this._popper.destroy(),this._popper=null))},this.tip,e),this._hoverState=""}update(){null!==this._popper&&this._popper.update()}isWithContent(){return Boolean(this.getTitle())}getTipElement(){if(this.tip)return this.tip;const t=document.createElement("div");return t.innerHTML=this._config.template,this.tip=t.children[0],this.tip}setContent(){const e=this.getTipElement();this.setElementContent(t.findOne(".tooltip-inner",e),this.getTitle()),e.classList.remove("fade","show")}setElementContent(t,e){if(null!==t)return r(e)?(e=a(e),void(this._config.html?e.parentNode!==t&&(t.innerHTML="",t.appendChild(e)):t.textContent=e.textContent)):void(this._config.html?(this._config.sanitize&&(e=Re(e,this._config.allowList,this._config.sanitizeFn)),t.innerHTML=e):t.textContent=e)}getTitle(){let t=this._element.getAttribute("data-bs-original-title");return t||(t="function"==typeof this._config.title?this._config.title.call(this._element):this._config.title),t}updateAttachment(t){return"right"===t?"end":"left"===t?"start":t}_initializeOnDelegatedTarget(t,e){const i=this.constructor.DATA_KEY;return(e=e||R.get(t.delegateTarget,i))||(e=new this.constructor(t.delegateTarget,this._getDelegateConfig()),R.set(t.delegateTarget,i,e)),e}_getOffset(){const{offset:t}=this._config;return"string"==typeof t?t.split(",").map(t=>Number.parseInt(t,10)):"function"==typeof t?e=>t(e,this._element):t}_getPopperConfig(t){const e={placement:t,modifiers:[{name:"flip",options:{fallbackPlacements:this._config.fallbackPlacements}},{name:"offset",options:{offset:this._getOffset()}},{name:"preventOverflow",options:{boundary:this._config.boundary}},{name:"arrow",options:{element:`.${this.constructor.NAME}-arrow`}},{name:"onChange",enabled:!0,phase:"afterWrite",fn:t=>this._handlePopperPlacementChange(t)}],onFirstUpdate:t=>{t.options.placement!==t.placement&&this._handlePopperPlacementChange(t)}};return{...e,..."function"==typeof this._config.popperConfig?this._config.popperConfig(e):this._config.popperConfig}}_addAttachmentClass(t){this.getTipElement().classList.add("bs-tooltip-"+this.updateAttachment(t))}_getAttachment(t){return ze[t.toUpperCase()]}_setListeners(){this._config.trigger.split(" ").forEach(t=>{if("click"===t)P.on(this._element,this.constructor.Event.CLICK,this._config.selector,t=>this.toggle(t));else if("manual"!==t){const e="hover"===t?this.constructor.Event.MOUSEENTER:this.constructor.Event.FOCUSIN,i="hover"===t?this.constructor.Event.MOUSELEAVE:this.constructor.Event.FOCUSOUT;P.on(this._element,e,this._config.selector,t=>this._enter(t)),P.on(this._element,i,this._config.selector,t=>this._leave(t))}}),this._hideModalHandler=()=>{this._element&&this.hide()},P.on(this._element.closest(".modal"),"hide.bs.modal",this._hideModalHandler),this._config.selector?this._config={...this._config,trigger:"manual",selector:""}:this._fixTitle()}_fixTitle(){const t=this._element.getAttribute("title"),e=typeof this._element.getAttribute("data-bs-original-title");(t||"string"!==e)&&(this._element.setAttribute("data-bs-original-title",t||""),!t||this._element.getAttribute("aria-label")||this._element.textContent||this._element.setAttribute("aria-label",t),this._element.setAttribute("title",""))}_enter(t,e){e=this._initializeOnDelegatedTarget(t,e),t&&(e._activeTrigger["focusin"===t.type?"focus":"hover"]=!0),e.getTipElement().classList.contains("show")||"show"===e._hoverState?e._hoverState="show":(clearTimeout(e._timeout),e._hoverState="show",e._config.delay&&e._config.delay.show?e._timeout=setTimeout(()=>{"show"===e._hoverState&&e.show()},e._config.delay.show):e.show())}_leave(t,e){e=this._initializeOnDelegatedTarget(t,e),t&&(e._activeTrigger["focusout"===t.type?"focus":"hover"]=e._element.contains(t.relatedTarget)),e._isWithActiveTrigger()||(clearTimeout(e._timeout),e._hoverState="out",e._config.delay&&e._config.delay.hide?e._timeout=setTimeout(()=>{"out"===e._hoverState&&e.hide()},e._config.delay.hide):e.hide())}_isWithActiveTrigger(){for(const t in this._activeTrigger)if(this._activeTrigger[t])return!0;return!1}_getConfig(t){const e=U.getDataAttributes(this._element);return Object.keys(e).forEach(t=>{We.has(t)&&delete e[t]}),(t={...this.constructor.Default,...e,..."object"==typeof t&&t?t:{}}).container=!1===t.container?document.body:a(t.container),"number"==typeof t.delay&&(t.delay={show:t.delay,hide:t.delay}),"number"==typeof t.title&&(t.title=t.title.toString()),"number"==typeof t.content&&(t.content=t.content.toString()),l("tooltip",t,this.constructor.DefaultType),t.sanitize&&(t.template=Re(t.template,t.allowList,t.sanitizeFn)),t}_getDelegateConfig(){const t={};if(this._config)for(const e in this._config)this.constructor.Default[e]!==this._config[e]&&(t[e]=this._config[e]);return t}_cleanTipClass(){const t=this.getTipElement(),e=t.getAttribute("class").match(Be);null!==e&&e.length>0&&e.map(t=>t.trim()).forEach(e=>t.classList.remove(e))}_handlePopperPlacementChange(t){const{state:e}=t;e&&(this.tip=e.elements.popper,this._cleanTipClass(),this._addAttachmentClass(this._getAttachment(e.placement)))}static jQueryInterface(t){return this.each((function(){const e=Fe.getOrCreateInstance(this,t);if("string"==typeof t){if(void 0===e[t])throw new TypeError(`No method named "${t}"`);e[t]()}}))}}_(Fe);const Ve=new RegExp("(^|\\s)bs-popover\\S+","g"),Ke={...Fe.Default,placement:"right",offset:[0,8],trigger:"click",content:"",template:'<div class="popover" role="tooltip"><div class="popover-arrow"></div><h3 class="popover-header"></h3><div class="popover-body"></div></div>'},Xe={...Fe.DefaultType,content:"(string|element|function)"},Ye={HIDE:"hide.bs.popover",HIDDEN:"hidden.bs.popover",SHOW:"show.bs.popover",SHOWN:"shown.bs.popover",INSERTED:"inserted.bs.popover",CLICK:"click.bs.popover",FOCUSIN:"focusin.bs.popover",FOCUSOUT:"focusout.bs.popover",MOUSEENTER:"mouseenter.bs.popover",MOUSELEAVE:"mouseleave.bs.popover"};class Qe extends Fe{static get Default(){return Ke}static get NAME(){return"popover"}static get Event(){return Ye}static get DefaultType(){return Xe}isWithContent(){return this.getTitle()||this._getContent()}getTipElement(){return this.tip||(this.tip=super.getTipElement(),this.getTitle()||t.findOne(".popover-header",this.tip).remove(),this._getContent()||t.findOne(".popover-body",this.tip).remove()),this.tip}setContent(){const e=this.getTipElement();this.setElementContent(t.findOne(".popover-header",e),this.getTitle());let i=this._getContent();"function"==typeof i&&(i=i.call(this._element)),this.setElementContent(t.findOne(".popover-body",e),i),e.classList.remove("fade","show")}_addAttachmentClass(t){this.getTipElement().classList.add("bs-popover-"+this.updateAttachment(t))}_getContent(){return this._element.getAttribute("data-bs-content")||this._config.content}_cleanTipClass(){const t=this.getTipElement(),e=t.getAttribute("class").match(Ve);null!==e&&e.length>0&&e.map(t=>t.trim()).forEach(e=>t.classList.remove(e))}static jQueryInterface(t){return this.each((function(){const e=Qe.getOrCreateInstance(this,t);if("string"==typeof t){if(void 0===e[t])throw new TypeError(`No method named "${t}"`);e[t]()}}))}}_(Qe);const Ge={offset:10,method:"auto",target:""},Ze={offset:"number",method:"string",target:"(string|element)"};class Je extends B{constructor(t,e){super(t),this._scrollElement="BODY"===this._element.tagName?window:this._element,this._config=this._getConfig(e),this._selector=`${this._config.target} .nav-link, ${this._config.target} .list-group-item, ${this._config.target} .dropdown-item`,this._offsets=[],this._targets=[],this._activeTarget=null,this._scrollHeight=0,P.on(this._scrollElement,"scroll.bs.scrollspy",()=>this._process()),this.refresh(),this._process()}static get Default(){return Ge}static get NAME(){return"scrollspy"}refresh(){const e=this._scrollElement===this._scrollElement.window?"offset":"position",i="auto"===this._config.method?e:this._config.method,s="position"===i?this._getScrollTop():0;this._offsets=[],this._targets=[],this._scrollHeight=this._getScrollHeight(),t.find(this._selector).map(e=>{const o=n(e),r=o?t.findOne(o):null;if(r){const t=r.getBoundingClientRect();if(t.width||t.height)return[U[i](r).top+s,o]}return null}).filter(t=>t).sort((t,e)=>t[0]-e[0]).forEach(t=>{this._offsets.push(t[0]),this._targets.push(t[1])})}dispose(){P.off(this._scrollElement,".bs.scrollspy"),super.dispose()}_getConfig(t){if("string"!=typeof(t={...Ge,...U.getDataAttributes(this._element),..."object"==typeof t&&t?t:{}}).target&&r(t.target)){let{id:i}=t.target;i||(i=e("scrollspy"),t.target.id=i),t.target="#"+i}return l("scrollspy",t,Ze),t}_getScrollTop(){return this._scrollElement===window?this._scrollElement.pageYOffset:this._scrollElement.scrollTop}_getScrollHeight(){return this._scrollElement.scrollHeight||Math.max(document.body.scrollHeight,document.documentElement.scrollHeight)}_getOffsetHeight(){return this._scrollElement===window?window.innerHeight:this._scrollElement.getBoundingClientRect().height}_process(){const t=this._getScrollTop()+this._config.offset,e=this._getScrollHeight(),i=this._config.offset+e-this._getOffsetHeight();if(this._scrollHeight!==e&&this.refresh(),t>=i){const t=this._targets[this._targets.length-1];this._activeTarget!==t&&this._activate(t)}else{if(this._activeTarget&&t<this._offsets[0]&&this._offsets[0]>0)return this._activeTarget=null,void this._clear();for(let e=this._offsets.length;e--;)this._activeTarget!==this._targets[e]&&t>=this._offsets[e]&&(void 0===this._offsets[e+1]||t<this._offsets[e+1])&&this._activate(this._targets[e])}}_activate(e){this._activeTarget=e,this._clear();const i=this._selector.split(",").map(t=>`${t}[data-bs-target="${e}"],${t}[href="${e}"]`),n=t.findOne(i.join(","));n.classList.contains("dropdown-item")?(t.findOne(".dropdown-toggle",n.closest(".dropdown")).classList.add("active"),n.classList.add("active")):(n.classList.add("active"),t.parents(n,".nav, .list-group").forEach(e=>{t.prev(e,".nav-link, .list-group-item").forEach(t=>t.classList.add("active")),t.prev(e,".nav-item").forEach(e=>{t.children(e,".nav-link").forEach(t=>t.classList.add("active"))})})),P.trigger(this._scrollElement,"activate.bs.scrollspy",{relatedTarget:e})}_clear(){t.find(this._selector).filter(t=>t.classList.contains("active")).forEach(t=>t.classList.remove("active"))}static jQueryInterface(t){return this.each((function(){const e=Je.getOrCreateInstance(this,t);if("string"==typeof t){if(void 0===e[t])throw new TypeError(`No method named "${t}"`);e[t]()}}))}}P.on(window,"load.bs.scrollspy.data-api",()=>{t.find('[data-bs-spy="scroll"]').forEach(t=>new Je(t))}),_(Je);class ti extends B{static get NAME(){return"tab"}show(){if(this._element.parentNode&&this._element.parentNode.nodeType===Node.ELEMENT_NODE&&this._element.classList.contains("active"))return;let e;const i=s(this._element),n=this._element.closest(".nav, .list-group");if(n){const i="UL"===n.nodeName||"OL"===n.nodeName?":scope > li > .active":".active";e=t.find(i,n),e=e[e.length-1]}const o=e?P.trigger(e,"hide.bs.tab",{relatedTarget:this._element}):null;if(P.trigger(this._element,"show.bs.tab",{relatedTarget:e}).defaultPrevented||null!==o&&o.defaultPrevented)return;this._activate(this._element,n);const r=()=>{P.trigger(e,"hidden.bs.tab",{relatedTarget:this._element}),P.trigger(this._element,"shown.bs.tab",{relatedTarget:e})};i?this._activate(i,i.parentNode,r):r()}_activate(e,i,n){const s=(!i||"UL"!==i.nodeName&&"OL"!==i.nodeName?t.children(i,".active"):t.find(":scope > li > .active",i))[0],o=n&&s&&s.classList.contains("fade"),r=()=>this._transitionComplete(e,s,n);s&&o?(s.classList.remove("show"),this._queueCallback(r,e,!0)):r()}_transitionComplete(e,i,n){if(i){i.classList.remove("active");const e=t.findOne(":scope > .dropdown-menu .active",i.parentNode);e&&e.classList.remove("active"),"tab"===i.getAttribute("role")&&i.setAttribute("aria-selected",!1)}e.classList.add("active"),"tab"===e.getAttribute("role")&&e.setAttribute("aria-selected",!0),f(e),e.classList.contains("fade")&&e.classList.add("show");let s=e.parentNode;if(s&&"LI"===s.nodeName&&(s=s.parentNode),s&&s.classList.contains("dropdown-menu")){const i=e.closest(".dropdown");i&&t.find(".dropdown-toggle",i).forEach(t=>t.classList.add("active")),e.setAttribute("aria-expanded",!0)}n&&n()}static jQueryInterface(t){return this.each((function(){const e=ti.getOrCreateInstance(this);if("string"==typeof t){if(void 0===e[t])throw new TypeError(`No method named "${t}"`);e[t]()}}))}}P.on(document,"click.bs.tab.data-api",'[data-bs-toggle="tab"], [data-bs-toggle="pill"], [data-bs-toggle="list"]',(function(t){["A","AREA"].includes(this.tagName)&&t.preventDefault(),h(this)||ti.getOrCreateInstance(this).show()})),_(ti);const ei={animation:"boolean",autohide:"boolean",delay:"number"},ii={animation:!0,autohide:!0,delay:5e3};class ni extends B{constructor(t,e){super(t),this._config=this._getConfig(e),this._timeout=null,this._hasMouseInteraction=!1,this._hasKeyboardInteraction=!1,this._setListeners()}static get DefaultType(){return ei}static get Default(){return ii}static get NAME(){return"toast"}show(){P.trigger(this._element,"show.bs.toast").defaultPrevented||(this._clearTimeout(),this._config.animation&&this._element.classList.add("fade"),this._element.classList.remove("hide"),f(this._element),this._element.classList.add("showing"),this._queueCallback(()=>{this._element.classList.remove("showing"),this._element.classList.add("show"),P.trigger(this._element,"shown.bs.toast"),this._maybeScheduleHide()},this._element,this._config.animation))}hide(){this._element.classList.contains("show")&&(P.trigger(this._element,"hide.bs.toast").defaultPrevented||(this._element.classList.remove("show"),this._queueCallback(()=>{this._element.classList.add("hide"),P.trigger(this._element,"hidden.bs.toast")},this._element,this._config.animation)))}dispose(){this._clearTimeout(),this._element.classList.contains("show")&&this._element.classList.remove("show"),super.dispose()}_getConfig(t){return t={...ii,...U.getDataAttributes(this._element),..."object"==typeof t&&t?t:{}},l("toast",t,this.constructor.DefaultType),t}_maybeScheduleHide(){this._config.autohide&&(this._hasMouseInteraction||this._hasKeyboardInteraction||(this._timeout=setTimeout(()=>{this.hide()},this._config.delay)))}_onInteraction(t,e){switch(t.type){case"mouseover":case"mouseout":this._hasMouseInteraction=e;break;case"focusin":case"focusout":this._hasKeyboardInteraction=e}if(e)return void this._clearTimeout();const i=t.relatedTarget;this._element===i||this._element.contains(i)||this._maybeScheduleHide()}_setListeners(){P.on(this._element,"click.dismiss.bs.toast",'[data-bs-dismiss="toast"]',()=>this.hide()),P.on(this._element,"mouseover.bs.toast",t=>this._onInteraction(t,!0)),P.on(this._element,"mouseout.bs.toast",t=>this._onInteraction(t,!1)),P.on(this._element,"focusin.bs.toast",t=>this._onInteraction(t,!0)),P.on(this._element,"focusout.bs.toast",t=>this._onInteraction(t,!1))}_clearTimeout(){clearTimeout(this._timeout),this._timeout=null}static jQueryInterface(t){return this.each((function(){const e=ni.getOrCreateInstance(this,t);if("string"==typeof t){if(void 0===e[t])throw new TypeError(`No method named "${t}"`);e[t](this)}}))}}return _(ni),{Alert:W,Button:q,Carousel:Z,Collapse:et,Dropdown:Ae,Modal:De,Offcanvas:Ne,Popover:Qe,ScrollSpy:Je,Tab:ti,Toast:ni,Tooltip:Fe}}));
+//# sourceMappingURL=bootstrap.bundle.min.js.map
\ No newline at end of file
From dea64ddad083d828aede004d42e926f4e4fd1015 Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Wed, 8 Jul 2026 01:52:22 +0200
Subject: [PATCH 24/43] Load skills into chat history
Switch the Skills catalog UI/API from scope-wide prompt pins to current-chat history loading, and remove user-facing removal of loaded skills.
Disable legacy active-skill prompt injection so forgotten scope config no longer inflates new-chat prompts.
---
.../message_loop_prompts_after/AGENTS.md | 2 +-
helpers/skills.py | 4 +-
helpers/skills.py.dox.md | 2 +
plugins/_skills/AGENTS.md | 15 +-
plugins/_skills/README.md | 23 ++-
plugins/_skills/api/skills_catalog.py | 93 ++++++----
plugins/_skills/plugin.yaml | 2 +-
.../prompts/agent.system.active_skills.md | 2 +-
plugins/_skills/webui/config-store.js | 139 ++-------------
plugins/_skills/webui/config.html | 36 +---
tests/test_skills_catalog_api.py | 163 ++++++++++++++++++
tests/test_skills_runtime.py | 29 ++++
12 files changed, 293 insertions(+), 217 deletions(-)
create mode 100644 tests/test_skills_catalog_api.py
diff --git a/extensions/python/message_loop_prompts_after/AGENTS.md b/extensions/python/message_loop_prompts_after/AGENTS.md
index ba8118a2f..cc80e0f34 100644
--- a/extensions/python/message_loop_prompts_after/AGENTS.md
+++ b/extensions/python/message_loop_prompts_after/AGENTS.md
@@ -7,9 +7,9 @@
## Ownership
- Ordered Python files own current datetime, skill recall/load context, agent info, parallel job status, and workdir extras injection.
-- Active skill instructions belong in prompt protocol.
- Explicitly loaded skill bodies belong in tool-result history with metadata so they can survive persistence and be reattached after compaction.
- Explicitly loaded skill IDs are chat-wide context data, not agent-local state.
+- Legacy active-skill prompt protocol injection must stay empty; selected skills are loaded through history.
## Local Contracts
diff --git a/helpers/skills.py b/helpers/skills.py
index 74488da9d..2e5d5c90a 100644
--- a/helpers/skills.py
+++ b/helpers/skills.py
@@ -1132,7 +1132,6 @@ def hide_chat_skill(agent: Agent, entry: Any) -> list[ActiveSkillEntry]:
CONTEXT_DATA_NAME_CHAT_VISIBLE_SKILLS,
visible_entries,
)
- unload_agent_skill(agent, normalized)
return get_hidden_skills(agent)
@@ -1183,8 +1182,7 @@ def clear_chat_skill_overrides(agent: Agent) -> list[ActiveSkillEntry]:
def build_active_skills_prompt(agent: Agent | None) -> str:
- items = _resolve_active_skill_entries(agent, get_active_skills(agent))
- return "\n\n".join(item["content"] for item in items if item.get("content")).strip()
+ return ""
def _format_skill_prompt(skill: Skill) -> str:
diff --git a/helpers/skills.py.dox.md b/helpers/skills.py.dox.md
index d269bb205..073997044 100644
--- a/helpers/skills.py.dox.md
+++ b/helpers/skills.py.dox.md
@@ -56,6 +56,8 @@
- Helper modules own reusable framework APIs and must preserve public callers unless all callers, tests, and docs are updated together.
- Update this file whenever public functions, classes, persistence behavior, path/security assumptions, side effects, or cross-module contracts change.
- Loaded skill names are chat-wide context data under `CONTEXT_DATA_NAME_LOADED_SKILLS`; legacy agent-local `loaded_skills` lists are migrated into context data and cleared when read.
+- Loaded skill bodies live in chat history; hiding a skill changes catalog visibility but does not remove the loaded-skill ledger.
+- `build_active_skills_prompt()` returns empty because selected skills are loaded through history, not prompt protocol.
- Invalid `SKILL.md` frontmatter emits a once-per-path scan warning with the skipped skill path/name and a line number when the parser can identify one directly.
- Observed side-effect areas: filesystem reads, filesystem deletion, plugin state, settings/state persistence, context data, secret handling.
- Imported dependency areas include: `__future__`, `dataclasses`, `helpers`, `os`, `pathlib`, `re`, `typing`.
diff --git a/plugins/_skills/AGENTS.md b/plugins/_skills/AGENTS.md
index ebb1af367..a1ca0d81e 100644
--- a/plugins/_skills/AGENTS.md
+++ b/plugins/_skills/AGENTS.md
@@ -2,25 +2,26 @@
## Purpose
-- Own active and hidden skill configuration injected into prompt protocol on each turn.
+- Own current-chat skill loading and hidden skill configuration.
## Ownership
-- `hooks.py` owns skill prompt injection and plugin lifecycle behavior.
-- `api/skills_catalog.py` owns skill catalog access.
-- `prompts/agent.system.active_skills.md` owns injected active-skill prompt content.
+- `hooks.py` owns skill config normalization.
+- `api/skills_catalog.py` owns skill catalog access and loading selected skills into chat history.
+- `prompts/agent.system.active_skills.md` is retained only for legacy prompt-protocol compatibility.
- `webui/` owns skill settings UI and store.
- `default_config.yaml`, `plugin.yaml`, `README.md`, and `LICENSE` own defaults, metadata, docs, and license.
## Local Contracts
-- Keep active skill lists bounded by configured caps.
+- Skills selected in `webui/` load into the current chat history only; do not store them as scope defaults.
+- Loaded skills are append-only from the user UI because their instructions live in chat history.
- Store configured skills in normalized portable paths.
-- Hidden skills affect catalog/search/load visibility but must not be injected as active prompt content.
+- Hidden skills affect catalog/search/load visibility but must not remove loaded skill history.
## Work Guidance
-- Coordinate active-skill resolution changes with core skill loading and settings UI.
+- Coordinate skill loading changes with `skills_tool`, loaded-skill history reattachment, and settings UI.
## Verification
diff --git a/plugins/_skills/README.md b/plugins/_skills/README.md
index 4d27c4f96..66ecbab2a 100644
--- a/plugins/_skills/README.md
+++ b/plugins/_skills/README.md
@@ -1,29 +1,28 @@
# Skills
-Skills is a built-in Agent Zero plugin that manages active skills across scope defaults and the current chat.
+Skills is a built-in Agent Zero plugin that manages skill loading and visibility for the current chat.
## What It Does
-- pins default skills for the current plugin scope
+- loads selected skills into current-chat history
- hides noisy skills from the model-facing available catalog, skill search, and load access
-- injects the effective active skills into prompt protocol on every turn
-- extends the same config screen with a current-chat mode so users can activate or hide skills live per conversation
+- shows loaded skills without offering removal, because loaded skill bodies are part of chat history
+- lets users hide or show skills live per conversation
- supports global and project scoped configurations without agent-profile variants
- links directly to the built-in Skills list
- links directly to the active project's Skills section when a project is active
## Why This Exists
-Agent Zero already supports loading skills dynamically with `skills_tool`, and already has great built-in skill management surfaces. What it did not have was a lightweight way to make a few skills feel "always on" for a specific scope without modifying the core prompt system.
+Agent Zero already supports loading skills dynamically with `skills_tool`, and already has great built-in skill management surfaces. What it did not have was a lightweight way to use that same history-backed skill loading from the Skills screen.
Skills fills that gap as a bundled built-in plugin.
-The shared active-skill state and prompt-resolution logic live in `helpers/skills.py`, and this plugin focuses on configuration, UI, and prompt injection.
+The shared skill discovery and loaded-skill ledger live in `helpers/skills.py`, and this plugin focuses on catalog UI, chat loading, and visibility.
## Notes
-- keep the active list short because every active skill is injected into prompt protocol every turn
-- the default cap is 20 active skills, and it can be raised or lowered in Skills plugin config
-- hidden skills are not capped because they are stored as control data, not injected into the prompt
-- selected skills are stored in normalized `/a0/...` form so configs stay portable across development and Docker-style layouts
-- scope defaults can be hidden or supplemented per chat without creating a new conversation
-- if a configured skill is not visible in the current agent scope, it is skipped quietly instead of breaking the prompt build
+- selected skills are appended to current-chat history using the same metadata shape as `skills_tool`
+- loaded skills are not removable from the UI; future context compaction may reattach their bodies from the loaded-skill ledger
+- hidden skills are stored as control data, not injected into the prompt
+- hidden skill paths are stored in normalized `/a0/...` form so configs stay portable across development and Docker-style layouts
+- if a configured hidden skill is not visible in the current agent scope, it is skipped quietly instead of breaking catalog builds
diff --git a/plugins/_skills/api/skills_catalog.py b/plugins/_skills/api/skills_catalog.py
index 38ec23bb5..104de3332 100644
--- a/plugins/_skills/api/skills_catalog.py
+++ b/plugins/_skills/api/skills_catalog.py
@@ -37,17 +37,36 @@ class SkillsCatalog(ApiHandler):
def _activate(self, input: dict, *, context_id: str) -> dict[str, Any]:
context = self._require_context(context_id)
skill_entry = self._require_skill_entry(input)
- skills.activate_chat_skill(context.get_agent(), skill_entry)
+ agent = context.get_agent()
+ skill = self._resolve_catalog_skill(skill_entry, context=context)
+ skill_name = str(skill.get("name") or skill_entry.get("name") or "").strip()
+ if not skill_name:
+ raise ValueError("Skill name is required")
+
+ skill_path = str(skill.get("path") or skill_entry.get("path") or "").strip()
+ skills.add_loaded_skill_name(agent, skill_name)
+ if not self._visible_skill_loaded(agent, skill_name):
+ content = skills.load_skill_for_agent(skill_name=skill_name, agent=agent)
+ if content.startswith("Error:"):
+ raise ValueError(content)
+ agent.hist_add_tool_result(
+ "skills_tool",
+ content,
+ skill_instructions={
+ "name": skill_name,
+ "path": skill_path,
+ "source": "skills_page:load",
+ "content_included": True,
+ },
+ )
save_tmp_chat(context)
return self._build_state(context_id=context.id)
def _deactivate(self, input: dict, *, context_id: str) -> dict[str, Any]:
- context = self._require_context(context_id)
- skill_entry = self._require_skill_entry(input)
- skills.deactivate_chat_skill(context.get_agent(), skill_entry)
- skills.unload_agent_skill(context.get_agent(), skill_entry)
- save_tmp_chat(context)
- return self._build_state(context_id=context.id)
+ return {
+ "ok": False,
+ "error": "Loaded skills are kept in chat history and cannot be removed.",
+ }
def _hide(self, input: dict, *, context_id: str) -> dict[str, Any]:
context = self._require_context(context_id)
@@ -64,10 +83,10 @@ class SkillsCatalog(ApiHandler):
return self._build_state(context_id=context.id)
def _clear(self, *, context_id: str) -> dict[str, Any]:
- context = self._require_context(context_id)
- skills.clear_chat_skill_overrides(context.get_agent())
- save_tmp_chat(context)
- return self._build_state(context_id=context.id)
+ return {
+ "ok": False,
+ "error": "Loaded skills are kept in chat history and cannot be removed.",
+ }
def _build_state(
self,
@@ -87,26 +106,13 @@ class SkillsCatalog(ApiHandler):
str(skill.get("name") or "").strip().lower(): skill for skill in catalog
}
+ loaded_entries = skills.get_loaded_skill_entries(agent)
scope_entries = skills.get_scope_active_skills(agent)
scope_hidden_entries = skills.get_scope_hidden_skills(agent)
chat_entries = skills.get_chat_active_skills(context)
disabled_entries = skills.get_chat_disabled_skills(context)
visible_entries = skills.get_chat_visible_skills(context)
hidden_entries = skills.get_hidden_skills(agent)
- active_entries = self._merge_entries(
- skills.get_active_skills(agent),
- self._filter_hidden_entries(
- self._get_loaded_skill_entries(agent),
- hidden_entries,
- ),
- )
-
- scope_keys = {
- self._entry_key(entry) for entry in scope_entries if self._entry_key(entry)
- }
- chat_keys = {
- self._entry_key(entry) for entry in chat_entries if self._entry_key(entry)
- }
return {
"ok": True,
@@ -123,15 +129,9 @@ class SkillsCatalog(ApiHandler):
entry,
catalog_by_key,
catalog_by_name,
- state_source=(
- "Pinned + chat"
- if key in scope_keys and key in chat_keys
- else "Pinned default"
- if key in scope_keys
- else "Chat"
- ),
+ state_source="Loaded in chat history",
)
- for entry in active_entries
+ for entry in loaded_entries
if (key := self._entry_key(entry))
],
"scope_skills": [
@@ -139,7 +139,7 @@ class SkillsCatalog(ApiHandler):
entry,
catalog_by_key,
catalog_by_name,
- state_source="Pinned default",
+ state_source="Scope default",
)
for entry in scope_entries
],
@@ -253,11 +253,28 @@ class SkillsCatalog(ApiHandler):
def _entry_key(self, entry: dict[str, Any]) -> str:
return str(entry.get("path") or entry.get("name") or "").strip().lower()
- def _get_loaded_skill_entries(self, agent: Any | None) -> list[dict[str, str]]:
- if not agent:
- return []
+ def _visible_skill_loaded(self, agent: Any, skill_name: str) -> bool:
+ output = getattr(getattr(agent, "history", None), "output", None)
+ if not callable(output):
+ return False
+ return any(
+ skills.skill_instruction_name(message) == skill_name
+ for message in output()
+ )
- return skills.get_loaded_skill_entries(agent)
+ def _resolve_catalog_skill(
+ self,
+ entry: dict[str, Any],
+ *,
+ context: AgentContext,
+ ) -> dict[str, Any]:
+ agent = context.get_agent()
+ project_name = projects.get_context_project_name(context) or ""
+ catalog = skills.list_skill_catalog(project_name=project_name, agent=agent)
+ return next(
+ (item for item in catalog if self._entry_matches_any(entry, [item])),
+ entry,
+ )
def _merge_entries(
self,
diff --git a/plugins/_skills/plugin.yaml b/plugins/_skills/plugin.yaml
index 920d57b6c..f5361e385 100644
--- a/plugins/_skills/plugin.yaml
+++ b/plugins/_skills/plugin.yaml
@@ -1,6 +1,6 @@
name: _skills
title: Skills
-description: Pin skills into prompt protocol on every turn.
+description: Load skills into the current chat history.
version: 1.0.0
always_enabled: true
settings_sections:
diff --git a/plugins/_skills/prompts/agent.system.active_skills.md b/plugins/_skills/prompts/agent.system.active_skills.md
index a90a62075..919d8145b 100644
--- a/plugins/_skills/prompts/agent.system.active_skills.md
+++ b/plugins/_skills/prompts/agent.system.active_skills.md
@@ -1,5 +1,5 @@
## active skills
-The following skills were manually activated by the User.
+The following skills were explicitly activated for this chat.
Treat them as already loaded instructions and follow them when relevant.
{{skills}}
diff --git a/plugins/_skills/webui/config-store.js b/plugins/_skills/webui/config-store.js
index 12ce77e69..f90b895fa 100644
--- a/plugins/_skills/webui/config-store.js
+++ b/plugins/_skills/webui/config-store.js
@@ -1,27 +1,9 @@
import * as API from "/js/api.js";
import { store as markdownModalStore } from "/components/modals/markdown/markdown-store.js";
import { store as chatsStore } from "/components/sidebar/chats/chats-store.js";
-import {
- toastFrontendError,
- toastFrontendInfo,
-} from "/components/notifications/notification-store.js";
+import { toastFrontendError } from "/components/notifications/notification-store.js";
const CATALOG_API = "/plugins/_skills/skills_catalog";
-const MAX_ACTIVE_SKILLS_FALLBACK = 20;
-
-function normalizeMaxActiveSkills(value) {
- if (typeof value === "boolean") return MAX_ACTIVE_SKILLS_FALLBACK;
-
- const numeric = typeof value === "number"
- ? value
- : Number.parseInt(String(value ?? "").trim(), 10);
-
- if (!Number.isFinite(numeric) || numeric < 1) {
- return MAX_ACTIVE_SKILLS_FALLBACK;
- }
-
- return Math.floor(numeric);
-}
function normalizeEntry(entry) {
if (!entry) return null;
@@ -62,11 +44,8 @@ function entriesMatch(left, right) {
function ensureConfig(config) {
if (!config || typeof config !== "object") return;
- config.max_active_skills = normalizeMaxActiveSkills(config.max_active_skills);
- const activeSkills = Array.isArray(config.active_skills) ? config.active_skills : [];
const hiddenSkills = Array.isArray(config.hidden_skills) ? config.hidden_skills : [];
- config.active_skills = compactEntries(activeSkills, config.max_active_skills);
config.hidden_skills = compactEntries(hiddenSkills);
}
@@ -94,15 +73,13 @@ window.createSkillsConfigModel = (context, config) => ({
mutatingChat: false,
catalog: [],
search: "",
- maxActiveSkills: MAX_ACTIVE_SKILLS_FALLBACK,
selectedSkills: [],
hiddenSkills: [],
chatContextAvailable: false,
initDefaults() {
ensureConfig(config);
- this.maxActiveSkills = config.max_active_skills;
- this.selectedSkills = [...this.activeEntries];
+ this.selectedSkills = [];
this.hiddenSkills = [...this.hiddenEntries];
},
@@ -110,50 +87,11 @@ window.createSkillsConfigModel = (context, config) => ({
return context?.openOptions?.focus === "chat";
},
- get activeEntries() {
- ensureConfig(config);
- return config.active_skills;
- },
-
get hiddenEntries() {
ensureConfig(config);
return config.hidden_skills;
},
- get selectedCount() {
- return this.selectedSkills.length;
- },
-
- async applyMaxActiveSkills(value, { notify = false } = {}) {
- ensureConfig(config);
-
- const nextLimit = normalizeMaxActiveSkills(value);
- const previousLimit = this.maxActiveSkills;
- const previousCount = this.selectedSkills.length;
-
- config.max_active_skills = nextLimit;
- this.maxActiveSkills = nextLimit;
-
- if (previousCount > nextLimit) {
- this._setSelectedSkills(this.selectedSkills.slice(0, nextLimit));
-
- if (notify) {
- await toastFrontendInfo(
- `Trimmed ${previousCount - nextLimit} pinned skill${previousCount - nextLimit === 1 ? "" : "s"} to match the new cap of ${nextLimit}.`,
- "Skills"
- );
- }
- return;
- }
-
- if (notify && previousLimit !== nextLimit) {
- await toastFrontendInfo(
- `Pinned skill cap set to ${nextLimit}.`,
- "Skills"
- );
- }
- },
-
get catalogMap() {
const byKey = new Map();
for (const skill of this.catalog) {
@@ -188,15 +126,13 @@ window.createSkillsConfigModel = (context, config) => ({
},
pinnedSubtitle() {
- return this.isChatMode
- ? "These skills are currently pinned into the prompt for this chat."
- : "These skills are pinned by default in this scope.";
+ return "These skills are loaded into the current chat history.";
},
allSkillsSubtitle() {
return this.isChatMode
- ? "Check a skill to pin it for this chat. Use the eye control to hide or show it in this chat."
- : "Check a skill to pin it by default. Use the eye control to hide or show it in this scope.";
+ ? "Check a skill to load it into this chat. Use the eye control to hide or show it in this chat."
+ : "Check a skill to load it into the current chat. Use the eye control to hide or show it in this scope.";
},
hiddenStateLabel() {
@@ -223,7 +159,7 @@ window.createSkillsConfigModel = (context, config) => ({
},
isPinDisabled(skill) {
- return this.mutatingChat || (!this.isSelected(skill) && this.selectedCount >= this.maxActiveSkills);
+ return this.mutatingChat || !this.chatContextAvailable || this.isSelected(skill);
},
isVisibilityDisabled() {
@@ -260,7 +196,7 @@ window.createSkillsConfigModel = (context, config) => ({
return name ? this.catalogMap.get(name) || null : null;
},
- _setSelectedSkills(entries, { writeConfig = true } = {}) {
+ _setSelectedSkills(entries) {
const normalized = [];
const seen = new Set();
@@ -270,13 +206,9 @@ window.createSkillsConfigModel = (context, config) => ({
if (!item || !key || seen.has(key)) continue;
seen.add(key);
normalized.push(item);
- if (normalized.length >= this.maxActiveSkills) break;
}
this.selectedSkills = normalized;
- if (writeConfig) {
- config.active_skills = compactEntries(normalized, this.maxActiveSkills);
- }
},
_setHiddenSkills(entries, { writeConfig = true } = {}) {
@@ -313,54 +245,28 @@ window.createSkillsConfigModel = (context, config) => ({
},
async togglePinnedSkill(skill, selected) {
- const nextEntries = this.selectedSkills.filter((entry) => !entriesMatch(entry, skill));
-
- if (selected) {
- if (this.selectedCount >= this.maxActiveSkills && !this.isSelected(skill)) {
- await toastFrontendInfo(
- `You can activate at most ${this.maxActiveSkills} skills.`,
- "Skills"
- );
- return;
- }
-
- nextEntries.push({
- name: String(skill.name || "").trim(),
- path: String(skill.path || "").trim(),
- });
+ if (!selected || this.isSelected(skill)) {
+ await this.loadCatalog();
+ return;
}
- this._setSelectedSkills(nextEntries, { writeConfig: !this.isChatMode });
-
- if (this.isChatMode && this.chatContextAvailable) {
- await this.submitChatAction(selected ? "activate" : "deactivate", skill);
+ if (!this.chatContextAvailable) {
+ await toastFrontendError("Open a chat before loading a skill.", "Skills");
+ await this.loadCatalog();
+ return;
}
- },
- async removeEntry(entry) {
- await this.togglePinnedSkill(entry, false);
- },
-
- async clearSelections() {
- const previous = [...this.selectedSkills];
- this._setSelectedSkills([], { writeConfig: !this.isChatMode });
- if (this.isChatMode && this.chatContextAvailable) {
- for (const entry of previous) {
- await this.submitChatAction("deactivate", entry);
- }
- }
+ await this.submitChatAction("activate", skill);
},
applyCatalogState(response) {
this.chatContextAvailable = !!response?.context_available;
const activeFromChat = Array.isArray(response?.active_skills) ? response.active_skills : null;
- const activeFromConfig = this.activeEntries;
const hiddenFromChat = Array.isArray(response?.hidden_skills) ? response.hidden_skills : null;
const hiddenFromConfig = this.hiddenEntries;
this._setSelectedSkills(
- this.isChatMode ? activeFromChat || [] : activeFromConfig,
- { writeConfig: !this.isChatMode },
+ activeFromChat || [],
);
this._setHiddenSkills(
this.isChatMode ? hiddenFromChat || [] : hiddenFromConfig,
@@ -374,7 +280,7 @@ window.createSkillsConfigModel = (context, config) => ({
const response = await API.callJsonApi(CATALOG_API, {
action: "list",
project_name: context.projectName || "",
- context_id: this.isChatMode ? chatsStore.selectedContext?.id || "" : "",
+ context_id: chatsStore.selectedContext?.id || "",
});
if (!response?.ok) {
@@ -382,17 +288,12 @@ window.createSkillsConfigModel = (context, config) => ({
}
this.catalog = Array.isArray(response.skills) ? response.skills : [];
- this.maxActiveSkills = normalizeMaxActiveSkills(response.max_active_skills);
- if (!this.isChatMode) {
- config.max_active_skills = this.maxActiveSkills;
- }
this.applyCatalogState(response);
} catch (error) {
this.catalog = [];
ensureConfig(config);
- this.maxActiveSkills = config.max_active_skills;
this.chatContextAvailable = false;
- this._setSelectedSkills(this.activeEntries);
+ this._setSelectedSkills([]);
this._setHiddenSkills(this.hiddenEntries);
await toastFrontendError(error?.message || "Failed to load skills", "Skills");
} finally {
@@ -422,10 +323,6 @@ window.createSkillsConfigModel = (context, config) => ({
}
this.catalog = Array.isArray(response.skills) ? response.skills : this.catalog;
- this.maxActiveSkills = normalizeMaxActiveSkills(response.max_active_skills);
- if (!this.isChatMode) {
- config.max_active_skills = this.maxActiveSkills;
- }
this.chatContextAvailable = !!response.context_available;
this.applyCatalogState(response);
return true;
diff --git a/plugins/_skills/webui/config.html b/plugins/_skills/webui/config.html
index 7885368ae..6965c0e18 100644
--- a/plugins/_skills/webui/config.html
+++ b/plugins/_skills/webui/config.html
@@ -15,26 +15,6 @@
<div class="skills-layout">
<div class="section-title">Skills</div>
- <template x-if="!isChatMode">
- <div class="field">
- <div class="field-label">
- <div class="field-title">Pinned skill cap</div>
- <div class="field-description">
- Maximum number of pinned skills stored in this scope. Higher values inject more instructions into every prompt.
- </div>
- </div>
- <div class="field-control">
- <input
- type="number"
- min="1"
- step="1"
- x-model.number="config.max_active_skills"
- @change="applyMaxActiveSkills(config.max_active_skills, { notify: true })"
- >
- </div>
- </div>
- </template>
-
<div class="skills-toolbar">
<label class="skills-search">
<span class="material-symbols-outlined">search</span>
@@ -54,11 +34,11 @@
</div>
<div class="skills-section">
- <div class="skills-panel-title">Pinned skills</div>
+ <div class="skills-panel-title">Loaded skills</div>
<div class="skills-panel-subtitle" x-text="pinnedSubtitle()"></div>
<template x-if="!loadingCatalog && selectedSkills.length === 0">
- <div class="skills-empty">No pinned skills yet. Pin from the list below.</div>
+ <div class="skills-empty">No loaded skills yet.</div>
</template>
<div class="skills-selected-list">
@@ -78,16 +58,6 @@
>
<span class="icon material-symbols-outlined">article</span>
</button>
- <button
- type="button"
- class="button cancel icon-button"
- title="Remove"
- aria-label="Remove skill"
- @click="$confirmClick($event, () => removeEntry(entry))"
- :disabled="mutatingChat"
- >
- <span class="icon material-symbols-outlined">close</span>
- </button>
</div>
</div>
</template>
@@ -118,7 +88,7 @@
'is-hidden': isHidden(skill),
}"
>
- <label class="skills-checkbox" :title="isSelected(skill) ? 'Pinned' : 'Pin skill'">
+ <label class="skills-checkbox" :title="isSelected(skill) ? 'Loaded in this chat' : 'Load skill into this chat'">
<input
type="checkbox"
:checked="isSelected(skill)"
diff --git a/tests/test_skills_catalog_api.py b/tests/test_skills_catalog_api.py
new file mode 100644
index 000000000..6dfaa1fd2
--- /dev/null
+++ b/tests/test_skills_catalog_api.py
@@ -0,0 +1,163 @@
+import asyncio
+import sys
+from pathlib import Path
+from types import SimpleNamespace
+
+PROJECT_ROOT = Path(__file__).resolve().parents[1]
+if str(PROJECT_ROOT) not in sys.path:
+ sys.path.insert(0, str(PROJECT_ROOT))
+
+from plugins._skills.api import skills_catalog
+
+
+class FakeHistory:
+ def __init__(self):
+ self.messages = []
+
+ def output(self):
+ return self.messages
+
+
+class FakeContext:
+ def __init__(self):
+ self.id = "ctx"
+ self.data = {}
+ self.agent = FakeAgent(self)
+
+ def get_agent(self):
+ return self.agent
+
+ def get_data(self, key, recursive=True):
+ return self.data.get(key)
+
+ def set_data(self, key, value, recursive=True):
+ self.data[key] = value
+
+
+class FakeAgent:
+ def __init__(self, context):
+ self.context = context
+ self.history = FakeHistory()
+ self.tool_results = []
+
+ def hist_add_tool_result(self, tool_name, tool_result, **kwargs):
+ content = {"tool_name": tool_name, "tool_result": tool_result, **kwargs}
+ message = {"ai": False, "content": content}
+ self.tool_results.append(content)
+ self.history.messages.append(message)
+ return SimpleNamespace(output=lambda: [message])
+
+
+def _patch_catalog(monkeypatch, context):
+ skill = {
+ "name": "demo-skill",
+ "description": "Demo skill.",
+ "path": "/a0/skills/demo-skill",
+ "origin": "Built-in",
+ "hidden": False,
+ }
+
+ monkeypatch.setattr(
+ skills_catalog.AgentContext,
+ "get",
+ staticmethod(lambda context_id: context if context_id == context.id else None),
+ )
+ monkeypatch.setattr(
+ skills_catalog.projects,
+ "get_context_project_name",
+ lambda _context: "",
+ )
+ monkeypatch.setattr(
+ skills_catalog.skills,
+ "list_skill_catalog",
+ lambda *args, **kwargs: [skill],
+ )
+ monkeypatch.setattr(
+ skills_catalog.skills,
+ "load_skill_for_agent",
+ lambda skill_name, agent: f"Skill: {skill_name}\n\nInstructions:\nUse it.",
+ )
+ monkeypatch.setattr(
+ skills_catalog.skills,
+ "add_loaded_skill_name",
+ lambda agent, skill_name: agent.context.set_data("loaded_skills", [skill_name]),
+ )
+ monkeypatch.setattr(
+ skills_catalog.skills,
+ "get_loaded_skill_entries",
+ lambda agent: [
+ {"name": name} for name in (agent.context.get_data("loaded_skills") or [])
+ ] if agent else [],
+ )
+ monkeypatch.setattr(skills_catalog.skills, "get_scope_active_skills", lambda agent: [])
+ monkeypatch.setattr(skills_catalog.skills, "get_scope_hidden_skills", lambda agent: [])
+ monkeypatch.setattr(skills_catalog.skills, "get_chat_active_skills", lambda context: [])
+ monkeypatch.setattr(skills_catalog.skills, "get_chat_disabled_skills", lambda context: [])
+ monkeypatch.setattr(skills_catalog.skills, "get_chat_visible_skills", lambda context: [])
+ monkeypatch.setattr(skills_catalog.skills, "get_hidden_skills", lambda agent: [])
+ monkeypatch.setattr(skills_catalog.skills, "get_max_active_skills", lambda **kwargs: 20)
+
+ saved = []
+ monkeypatch.setattr(skills_catalog, "save_tmp_chat", lambda ctx: saved.append(ctx.id))
+ return saved
+
+
+def test_skills_catalog_activate_loads_skill_into_chat_history(monkeypatch):
+ context = FakeContext()
+ saved = _patch_catalog(monkeypatch, context)
+ handler = skills_catalog.SkillsCatalog(None, None)
+
+ response = asyncio.run(
+ handler.process(
+ {
+ "action": "activate",
+ "context_id": "ctx",
+ "skill": {"name": "demo-skill", "path": "/a0/skills/demo-skill"},
+ },
+ None,
+ )
+ )
+
+ assert response["ok"] is True, response
+ assert context.get_data("loaded_skills") == ["demo-skill"]
+ assert len(context.agent.tool_results) == 1
+ assert "Skill: demo-skill" in context.agent.tool_results[0]["tool_result"]
+ assert response["active_skills"][0]["state_source"] == "Loaded in chat history"
+ assert saved == ["ctx"]
+
+ duplicate = asyncio.run(
+ handler.process(
+ {
+ "action": "activate",
+ "context_id": "ctx",
+ "skill": {"name": "demo-skill", "path": "/a0/skills/demo-skill"},
+ },
+ None,
+ )
+ )
+
+ assert duplicate["ok"] is True
+ assert len(context.agent.tool_results) == 1
+
+
+def test_skills_catalog_deactivate_does_not_remove_loaded_skill(monkeypatch):
+ context = FakeContext()
+ _patch_catalog(monkeypatch, context)
+ context.set_data("loaded_skills", ["demo-skill"])
+ handler = skills_catalog.SkillsCatalog(None, None)
+
+ response = asyncio.run(
+ handler.process(
+ {
+ "action": "deactivate",
+ "context_id": "ctx",
+ "skill": {"name": "demo-skill"},
+ },
+ None,
+ )
+ )
+
+ assert response["ok"] is False
+ assert "cannot be removed" in response["error"]
+ assert context.get_data("loaded_skills") == ["demo-skill"]
+ assert context.agent.tool_results == []
diff --git a/tests/test_skills_runtime.py b/tests/test_skills_runtime.py
index aa988ba2d..a66567bed 100644
--- a/tests/test_skills_runtime.py
+++ b/tests/test_skills_runtime.py
@@ -150,6 +150,35 @@ def test_hidden_skills_are_not_capped_like_active_skills():
assert len(runtime.get_hidden_skills(agent)) == 25
+def test_active_skill_prompt_protocol_is_disabled(monkeypatch):
+ monkeypatch.setattr(
+ runtime.plugin_helpers,
+ "get_plugin_config",
+ lambda *args, **kwargs: _scope_config([{"name": "Pinned"}]),
+ )
+ agent = DummyAgent()
+
+ assert runtime.get_active_skills(agent) == [{"name": "Pinned"}]
+ assert runtime.build_active_skills_prompt(agent) == ""
+
+
+def test_hiding_skill_does_not_unload_history_loaded_skill():
+ agent = DummyAgent()
+ agent.context.set_data(
+ runtime.CONTEXT_DATA_NAME_LOADED_SKILLS,
+ ["history-skill"],
+ )
+
+ runtime.hide_chat_skill(agent, {"name": "history-skill"})
+
+ assert agent.context.get_data(runtime.CONTEXT_DATA_NAME_LOADED_SKILLS) == [
+ "history-skill"
+ ]
+ assert runtime.get_chat_disabled_skills(agent.context) == [
+ {"name": "history-skill"}
+ ]
+
+
def test_chat_activation_can_override_scope_defaults(monkeypatch):
monkeypatch.setattr(
runtime.plugin_helpers,
From 0de0fcec0e3063d1d4ce3d7eec1ace6c96d972e7 Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Wed, 8 Jul 2026 02:03:00 +0200
Subject: [PATCH 25/43] Simplify relevant skill recall
Search the raw user message when recalling relevant skills instead of the rendered history wrapper.
Replace the stopword catalog with structural matching: names use normal terms, tags/triggers use long terms or phrase matches, and descriptions require phrase matches.
---
.../message_loop_prompts_after/AGENTS.md | 1 +
.../_63_recall_relevant_skills.py | 10 ++--
helpers/skills.py | 15 +++---
helpers/skills.py.dox.md | 1 +
tests/test_skills_runtime.py | 21 ++++++++
tests/test_tool_action_contracts.py | 51 +++++++++++++++++++
6 files changed, 90 insertions(+), 9 deletions(-)
diff --git a/extensions/python/message_loop_prompts_after/AGENTS.md b/extensions/python/message_loop_prompts_after/AGENTS.md
index cc80e0f34..a03541caf 100644
--- a/extensions/python/message_loop_prompts_after/AGENTS.md
+++ b/extensions/python/message_loop_prompts_after/AGENTS.md
@@ -16,6 +16,7 @@
- Keep injected content bounded and clearly attributed.
- Preserve ordering where later prompt extras depend on earlier recall or load results.
- Do not expose secrets or private files from workdir extras.
+- Relevant-skill recall should search the raw user message when available, not the rendered history wrapper.
## Work Guidance
diff --git a/extensions/python/message_loop_prompts_after/_63_recall_relevant_skills.py b/extensions/python/message_loop_prompts_after/_63_recall_relevant_skills.py
index c4b2a45e8..a810cece4 100644
--- a/extensions/python/message_loop_prompts_after/_63_recall_relevant_skills.py
+++ b/extensions/python/message_loop_prompts_after/_63_recall_relevant_skills.py
@@ -8,9 +8,13 @@ class RecallRelevantSkills(Extension):
if not self.agent or loop_data.iteration != 0:
return
- user_instruction = (
- loop_data.user_message.output_text() if loop_data.user_message else ""
- ).strip()
+ content = loop_data.user_message.content if loop_data.user_message else ""
+ if isinstance(content, dict):
+ user_instruction = str(content.get("user_message") or "").strip()
+ else:
+ user_instruction = (
+ loop_data.user_message.output_text() if loop_data.user_message else ""
+ ).strip()
if len(user_instruction) < 8:
return
diff --git a/helpers/skills.py b/helpers/skills.py
index 2e5d5c90a..1eeaa8504 100644
--- a/helpers/skills.py
+++ b/helpers/skills.py
@@ -543,11 +543,15 @@ def search_skills(
if not q:
return []
- raw_terms = [t for t in re.split(r"\s+", q) if t]
+ raw_terms = re.findall(r"[a-z0-9][a-z0-9_-]*", q)
terms = [
t for t in raw_terms
- if len(t) >= 3 or any(ch.isdigit() for ch in t)
- ] or raw_terms
+ if len(t) >= 4 or any(ch.isdigit() for ch in t)
+ ]
+ long_terms = [
+ t for t in raw_terms
+ if len(t) >= 6 or any(ch.isdigit() for ch in t)
+ ]
candidates = list_skills(agent, include_hidden=include_hidden)
scored: List[Tuple[int, Skill]] = []
@@ -568,14 +572,13 @@ def search_skills(
score += 4
if any(q in tag for tag in tags):
score += 3
- if any(q in trigger for trigger in triggers):
+ if any(q in trigger or trigger in q for trigger in triggers):
score += 8
for term in terms:
if term in name:
score += 3
- if term in desc:
- score += 2
+ for term in long_terms:
if any(term in tag for tag in tags):
score += 1
if any(term in trigger for trigger in triggers):
diff --git a/helpers/skills.py.dox.md b/helpers/skills.py.dox.md
index 073997044..902caf3fc 100644
--- a/helpers/skills.py.dox.md
+++ b/helpers/skills.py.dox.md
@@ -58,6 +58,7 @@
- Loaded skill names are chat-wide context data under `CONTEXT_DATA_NAME_LOADED_SKILLS`; legacy agent-local `loaded_skills` lists are migrated into context data and cleared when read.
- Loaded skill bodies live in chat history; hiding a skill changes catalog visibility but does not remove the loaded-skill ledger.
- `build_active_skills_prompt()` returns empty because selected skills are loaded through history, not prompt protocol.
+- `search_skills()` normalizes query words, scores normal terms against skill names, and scores only long terms against tags/triggers; descriptions match only full query phrases so generic prose does not produce irrelevant suggestions.
- Invalid `SKILL.md` frontmatter emits a once-per-path scan warning with the skipped skill path/name and a line number when the parser can identify one directly.
- Observed side-effect areas: filesystem reads, filesystem deletion, plugin state, settings/state persistence, context data, secret handling.
- Imported dependency areas include: `__future__`, `dataclasses`, `helpers`, `os`, `pathlib`, `re`, `typing`.
diff --git a/tests/test_skills_runtime.py b/tests/test_skills_runtime.py
index a66567bed..8228e4c36 100644
--- a/tests/test_skills_runtime.py
+++ b/tests/test_skills_runtime.py
@@ -473,6 +473,27 @@ def test_browser_skills_rank_for_browser_trigger_phrases(monkeypatch):
assert "browser-form-workflows" in form_results
+def test_skill_search_does_not_score_description_terms_alone(monkeypatch):
+ browser_automation = runtime.Skill(
+ name="browser-automation",
+ description=(
+ "Use for browser automation, screenshots, forms, uploads, "
+ "and complex tool workflows."
+ ),
+ path=Path("/skills/browser-automation"),
+ skill_md_path=Path("/skills/browser-automation/SKILL.md"),
+ )
+ monkeypatch.setattr(runtime, "list_skills", lambda *args, **kwargs: [browser_automation])
+
+ rendered_user_message = (
+ 'user: {"user_message": "Please reply with exactly OK. Do not use tools.", '
+ '"attachments": []}'
+ )
+
+ assert runtime.search_skills(rendered_user_message) == []
+ assert runtime.search_skills("Open a browser screenshot?", limit=1) == [browser_automation]
+
+
def test_host_computer_use_ranks_before_linux_desktop_for_host_screen_queries(monkeypatch):
host_skill = runtime.skill_from_markdown(
PROJECT_ROOT / "plugins" / "_a0_connector" / "skills" / "host-computer-use" / "SKILL.md"
diff --git a/tests/test_tool_action_contracts.py b/tests/test_tool_action_contracts.py
index ed21388cc..c85e83f69 100644
--- a/tests/test_tool_action_contracts.py
+++ b/tests/test_tool_action_contracts.py
@@ -140,6 +140,8 @@ def _load_skills_tool(monkeypatch, skill_root: Path):
skills_stub.set_loaded_skill_names = _set_loaded_skill_names
skills_stub.skill_instruction_name = _skill_instruction_name
monkeypatch.setitem(sys.modules, "helpers.skills", skills_stub)
+ import helpers
+ monkeypatch.setattr(helpers, "skills", skills_stub, raising=False)
print_style_stub = types.ModuleType("helpers.print_style")
print_style_stub.PrintStyle = lambda *args, **kwargs: types.SimpleNamespace(
@@ -213,6 +215,33 @@ def _load_loaded_skills_extension(monkeypatch, skill_root: Path):
return importlib.import_module(module_name)
+def _load_relevant_skills_extension(monkeypatch, queries: list[str]):
+ extension_stub = types.ModuleType("helpers.extension")
+ extension_stub.Extension = _FakeExtension
+ monkeypatch.setitem(sys.modules, "helpers.extension", extension_stub)
+
+ agent_stub = types.ModuleType("agent")
+ agent_stub.LoopData = lambda **kwargs: types.SimpleNamespace(**kwargs)
+ monkeypatch.setitem(sys.modules, "agent", agent_stub)
+
+ skills_stub = types.ModuleType("helpers.skills")
+
+ def _search_skills(query, *args, **kwargs):
+ queries.append(query)
+ return []
+
+ skills_stub.search_skills = _search_skills
+ monkeypatch.setitem(sys.modules, "helpers.skills", skills_stub)
+
+ import helpers
+
+ monkeypatch.setattr(helpers, "skills", skills_stub, raising=False)
+
+ module_name = "extensions.python.message_loop_prompts_after._63_recall_relevant_skills"
+ sys.modules.pop(module_name, None)
+ return importlib.import_module(module_name)
+
+
def _load_computer_use_remote_tool(monkeypatch):
_install_tool_stub(monkeypatch)
@@ -471,6 +500,28 @@ def test_loaded_skills_extension_keeps_reattachments_under_budget(
assert agent.added_tool_results == []
+def test_relevant_skill_recall_uses_raw_user_message(monkeypatch):
+ queries: list[str] = []
+ module = _load_relevant_skills_extension(monkeypatch, queries)
+ agent = types.SimpleNamespace()
+ loop_data = types.SimpleNamespace(
+ iteration=0,
+ user_message=types.SimpleNamespace(
+ content={
+ "system_message": [],
+ "user_message": "Open a browser and take a screenshot.",
+ "attachments": [],
+ },
+ output_text=lambda: 'user: {"user_message": "wrapped"}',
+ ),
+ extras_temporary={},
+ )
+
+ asyncio.run(module.RecallRelevantSkills(agent).execute(loop_data=loop_data))
+
+ assert queries == ["Open a browser and take a screenshot."]
+
+
def test_skills_tool_read_file_action_reads_inside_skill_dir(
monkeypatch, tmp_path: Path
):
From f8b06e0b12bb41fd4ca840906fe20b5a4da19e49 Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Wed, 8 Jul 2026 12:42:43 +0200
Subject: [PATCH 26/43] Remove legacy skill prompt cleanup
Delete the empty active-skills prompt extension and its unused prompt template now that selected skills are loaded through chat history instead of protocol. Stop clearing legacy loaded_skills keys from protocol/extras in the loaded-skills hook while keeping history reattachment for compacted skill bodies.
---
.../message_loop_prompts_after/AGENTS.md | 6 ++---
.../_65_include_loaded_skills.py | 3 ---
plugins/_skills/AGENTS.md | 1 -
.../_66_include_active_skills.py | 23 -------------------
.../prompts/agent.system.active_skills.md | 5 ----
tests/test_tool_action_contracts.py | 6 ++---
6 files changed, 5 insertions(+), 39 deletions(-)
delete mode 100644 plugins/_skills/extensions/python/message_loop_prompts_after/_66_include_active_skills.py
delete mode 100644 plugins/_skills/prompts/agent.system.active_skills.md
diff --git a/extensions/python/message_loop_prompts_after/AGENTS.md b/extensions/python/message_loop_prompts_after/AGENTS.md
index a03541caf..d18b1c5a1 100644
--- a/extensions/python/message_loop_prompts_after/AGENTS.md
+++ b/extensions/python/message_loop_prompts_after/AGENTS.md
@@ -2,14 +2,14 @@
## Purpose
-- Own prompt protocol and extras appended around primary message-loop prompt construction.
+- Own prompt protocol, prompt extras, and history reattachment around primary message-loop prompt construction.
## Ownership
-- Ordered Python files own current datetime, skill recall/load context, agent info, parallel job status, and workdir extras injection.
+- Ordered Python files own current datetime, relevant-skill hints, loaded-skill history reattachment, agent info, parallel job status, and workdir extras injection.
- Explicitly loaded skill bodies belong in tool-result history with metadata so they can survive persistence and be reattached after compaction.
- Explicitly loaded skill IDs are chat-wide context data, not agent-local state.
-- Legacy active-skill prompt protocol injection must stay empty; selected skills are loaded through history.
+- Skills must not write selected or loaded skill bodies into protocol or extras.
## Local Contracts
diff --git a/extensions/python/message_loop_prompts_after/_65_include_loaded_skills.py b/extensions/python/message_loop_prompts_after/_65_include_loaded_skills.py
index d7effd627..6c604005a 100644
--- a/extensions/python/message_loop_prompts_after/_65_include_loaded_skills.py
+++ b/extensions/python/message_loop_prompts_after/_65_include_loaded_skills.py
@@ -14,9 +14,6 @@ class IncludeLoadedSkills(Extension):
if not self.agent:
return
- loop_data.protocol_persistent.pop("loaded_skills", None)
- loop_data.extras_persistent.pop("loaded_skills", None)
-
skill_names = skills.get_loaded_skill_names(self.agent)
if not skill_names:
return
diff --git a/plugins/_skills/AGENTS.md b/plugins/_skills/AGENTS.md
index a1ca0d81e..eaddb8014 100644
--- a/plugins/_skills/AGENTS.md
+++ b/plugins/_skills/AGENTS.md
@@ -8,7 +8,6 @@
- `hooks.py` owns skill config normalization.
- `api/skills_catalog.py` owns skill catalog access and loading selected skills into chat history.
-- `prompts/agent.system.active_skills.md` is retained only for legacy prompt-protocol compatibility.
- `webui/` owns skill settings UI and store.
- `default_config.yaml`, `plugin.yaml`, `README.md`, and `LICENSE` own defaults, metadata, docs, and license.
diff --git a/plugins/_skills/extensions/python/message_loop_prompts_after/_66_include_active_skills.py b/plugins/_skills/extensions/python/message_loop_prompts_after/_66_include_active_skills.py
deleted file mode 100644
index ea9ced8b5..000000000
--- a/plugins/_skills/extensions/python/message_loop_prompts_after/_66_include_active_skills.py
+++ /dev/null
@@ -1,23 +0,0 @@
-from __future__ import annotations
-
-from helpers import skills
-from agent import LoopData
-from helpers.extension import Extension
-
-
-class IncludeActiveSkills(Extension):
- async def execute(self, loop_data: LoopData = LoopData(), **kwargs):
- if not self.agent:
- return
-
- protocol = loop_data.protocol_persistent
- protocol.pop("active_skills", None)
-
- content = skills.build_active_skills_prompt(self.agent)
- if not content:
- return
-
- protocol["active_skills"] = self.agent.read_prompt(
- "agent.system.active_skills.md",
- skills=content,
- )
diff --git a/plugins/_skills/prompts/agent.system.active_skills.md b/plugins/_skills/prompts/agent.system.active_skills.md
deleted file mode 100644
index 919d8145b..000000000
--- a/plugins/_skills/prompts/agent.system.active_skills.md
+++ /dev/null
@@ -1,5 +0,0 @@
-## active skills
-The following skills were explicitly activated for this chat.
-Treat them as already loaded instructions and follow them when relevant.
-
-{{skills}}
diff --git a/tests/test_tool_action_contracts.py b/tests/test_tool_action_contracts.py
index c85e83f69..a4cb49393 100644
--- a/tests/test_tool_action_contracts.py
+++ b/tests/test_tool_action_contracts.py
@@ -430,8 +430,8 @@ def test_loaded_skills_extension_reattaches_missing_body_after_compaction(
module = _load_loaded_skills_extension(monkeypatch, tmp_path)
agent = _FakeLoadedSkillAgent()
loop_data = types.SimpleNamespace(
- protocol_persistent={"loaded_skills": "legacy"},
- extras_persistent={"loaded_skills": "legacy"},
+ protocol_persistent={},
+ extras_persistent={},
history_output=[
{
"ai": False,
@@ -442,8 +442,6 @@ def test_loaded_skills_extension_reattaches_missing_body_after_compaction(
asyncio.run(module.IncludeLoadedSkills(agent).execute(loop_data))
- assert "loaded_skills" not in loop_data.protocol_persistent
- assert "loaded_skills" not in loop_data.extras_persistent
assert len(agent.added_tool_results) == 1
added = agent.added_tool_results[0]
assert added["tool_name"] == "skills_tool"
From bcf2634000dbe3655afae15686e1d0831036a213 Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Wed, 8 Jul 2026 14:29:15 +0200
Subject: [PATCH 27/43] Add AGENTS.md protocol guidance
Load active-project AGENTS.md path-chain guidance into the protocol area without duplicating the project root instructions.
Move the AGENTS.md protocol wording into a prompt template, reuse existing file/path helpers, and cover direct-path discovery plus prompt assembly with focused tests.
---
extensions/python/system_prompt/AGENTS.md | 2 +-
.../system_prompt/_14_project_prompt.py | 7 ++
helpers/AGENTS.md | 2 +-
helpers/projects.py | 84 +++++++++++++++++--
helpers/projects.py.dox.md | 4 +
prompts/agent.protocol.projects.agents_md.md | 6 ++
tests/test_projects.py | 62 ++++++++++++++
tests/test_prompt_protocol.py | 46 ++++++++++
8 files changed, 203 insertions(+), 10 deletions(-)
create mode 100644 prompts/agent.protocol.projects.agents_md.md
diff --git a/extensions/python/system_prompt/AGENTS.md b/extensions/python/system_prompt/AGENTS.md
index 34b90f78c..c4aa1bfd8 100644
--- a/extensions/python/system_prompt/AGENTS.md
+++ b/extensions/python/system_prompt/AGENTS.md
@@ -7,7 +7,7 @@
## Ownership
- Ordered Python files own main, tools, MCP, secrets, skills, and project prompt sections.
-- Active project instruction bodies are moved into prompt protocol; the system prompt keeps project metadata and stable project rules.
+- Active project instruction bodies and active-project AGENTS.md path-chain guidance are moved into prompt protocol; the system prompt keeps project metadata and stable project rules.
## Local Contracts
diff --git a/extensions/python/system_prompt/_14_project_prompt.py b/extensions/python/system_prompt/_14_project_prompt.py
index cd0363b4c..4afa620cd 100644
--- a/extensions/python/system_prompt/_14_project_prompt.py
+++ b/extensions/python/system_prompt/_14_project_prompt.py
@@ -25,9 +25,16 @@ async def build_prompt(agent: Agent, loop_data: LoopData | None = None) -> str:
result = agent.read_prompt("agent.system.projects.main.md")
project_name = agent.context.get_data(projects.CONTEXT_DATA_KEY_PROJECT)
if loop_data:
+ loop_data.protocol_persistent.pop("agents_md_instructions", None)
loop_data.protocol_persistent.pop("project_instructions", None)
if project_name:
project_vars = projects.build_system_prompt_vars(project_name)
+ if loop_data and project_vars.get("include_agents_md", True):
+ agents_md_protocol = projects.build_agents_md_protocol(project_name)
+ if agents_md_protocol:
+ loop_data.protocol_persistent["agents_md_instructions"] = (
+ agents_md_protocol
+ )
if loop_data and project_vars.get("project_instructions"):
loop_data.protocol_persistent["project_instructions"] = agent.read_prompt(
"agent.protocol.projects.instructions.md",
diff --git a/helpers/AGENTS.md b/helpers/AGENTS.md
index dc7584d32..39d24d71e 100644
--- a/helpers/AGENTS.md
+++ b/helpers/AGENTS.md
@@ -15,7 +15,7 @@
- Preserve public helper APIs used by core code and plugins unless all callers, docs, and tests are updated.
- Use structured parsers and serializers for YAML, JSON, paths, and URLs instead of ad hoc string handling.
- Keep path handling constrained to intended roots for user files, uploads, downloads, projects, and workdirs.
-- Project metadata defaults must remain backwards-compatible; missing `include_agents_md` is treated as enabled and project instruction file content is injected with an explicit source path.
+- Project metadata defaults must remain backwards-compatible; missing `include_agents_md` is treated as enabled, project instruction file content is injected with an explicit source path, and active-project AGENTS.md path-chain guidance is assembled into prompt protocol without duplicating the project root AGENTS.md.
- Do not hardcode secrets, provider keys, local absolute paths, or environment-specific values.
- Use `RepairableException` for errors an agent may be able to fix.
- This directory is a file-documented DOX profile: every direct `*.py` helper module must have a same-directory `*.py.dox.md` file named by appending `.dox.md` to the full Python filename.
diff --git a/helpers/projects.py b/helpers/projects.py
index f146d02a6..41bbf9d44 100644
--- a/helpers/projects.py
+++ b/helpers/projects.py
@@ -1,5 +1,5 @@
import os
-from typing import Literal, NotRequired, TypedDict, TYPE_CHECKING, cast
+from typing import NotRequired, TypedDict, TYPE_CHECKING, cast
from helpers import files, dirty_json, persist_chat, file_tree, extension
from helpers.print_style import PrintStyle
@@ -15,7 +15,13 @@ PROJECT_KNOWLEDGE_DIR = "knowledge"
PROJECT_SKILLS_DIR = "skills"
PROJECT_HEADER_FILE = "project.json"
PROJECT_MCP_SERVERS_FILE = "mcp_servers.json"
-PROJECT_AGENTS_MD_FILES = ("AGENTS.md", "Agents.md", "agents.md")
+PROJECT_AGENTS_MD_FILES = (
+ "AGENTS.override.md",
+ "AGENTS.Override.md",
+ "AGENTS.md",
+ "Agents.md",
+ "agents.md",
+)
DEFAULT_MCP_SERVERS_CONFIG = '{\n "mcpServers": {}\n}'
CONTEXT_DATA_KEY_PROJECT = "project"
@@ -466,9 +472,10 @@ def deactivate_project_in_chats(name: str):
def build_system_prompt_vars(name: str):
project_data = load_basic_project_data(name)
main_instructions = project_data.get("instructions", "") or ""
+ include_agents_md = project_data.get("include_agents_md", True)
instruction_files = get_project_instruction_files(
name,
- include_agents_md=project_data.get("include_agents_md", True),
+ include_agents_md=include_agents_md,
)
instruction_parts = [
main_instructions,
@@ -481,11 +488,75 @@ def build_system_prompt_vars(name: str):
"project_name": project_data.get("title", ""),
"project_description": project_data.get("description", ""),
"project_instructions": complete_instructions or "",
+ "include_agents_md": include_agents_md,
"project_path": files.normalize_a0_path(get_project_folder(name)),
"project_git_url": project_data.get("git_url", ""),
}
+def get_agents_md_chain(root: str, target: str) -> list[tuple[str, str]]:
+ root_real = os.path.realpath(files.fix_dev_path(root))
+ target_real = os.path.realpath(files.fix_dev_path(target))
+ if os.path.isfile(target_real):
+ target_real = os.path.dirname(target_real)
+
+ if files.is_in_dir(target_real, root_real):
+ dirs = []
+ cursor = target_real
+ while True:
+ dirs.append(cursor)
+ if cursor == root_real:
+ break
+ parent = os.path.dirname(cursor)
+ if parent == cursor:
+ break
+ cursor = parent
+ dirs.reverse()
+ else:
+ dirs = [root_real]
+
+ chain = []
+ for dir_path in dirs:
+ for filename in PROJECT_AGENTS_MD_FILES:
+ matches = files.read_text_files_in_dir(dir_path, pattern=filename)
+ if filename not in matches:
+ continue
+ chain.append((files.get_abs_path(dir_path, filename), matches[filename]))
+ break
+ return chain
+
+
+def build_agents_md_protocol(name: str, target: str | None = None) -> str:
+ project_folder = get_project_folder(name)
+ project_agents_md = get_project_agents_md_instruction_file(name)
+ project_agents_md_path = (
+ os.path.realpath(files.fix_dev_path(project_agents_md[0]))
+ if project_agents_md
+ else ""
+ )
+ entries = [
+ (path, content)
+ for path, content in get_agents_md_chain(
+ files.get_abs_path(""),
+ target or project_folder,
+ )
+ if os.path.realpath(path) != project_agents_md_path
+ ]
+ if not entries:
+ return ""
+
+ instructions = []
+ for path, content in entries:
+ instructions.append(
+ f"### path: {files.normalize_a0_path(path)}\n\n{content.strip()}"
+ )
+ return files.read_prompt_file(
+ "agent.protocol.projects.agents_md.md",
+ _directories=["prompts"],
+ agents_md_instructions="\n\n".join(instructions),
+ ).strip()
+
+
def get_additional_instructions_files(name: str):
instructions_folder = files.get_abs_path(
get_project_folder(name), PROJECT_META_DIR, PROJECT_INSTRUCTIONS_DIR
@@ -522,11 +593,8 @@ def get_project_instruction_files(
def get_project_agents_md_instruction_file(name: str) -> tuple[str, str] | None:
project_folder = get_project_folder(name)
- for filename in PROJECT_AGENTS_MD_FILES:
- matches = files.read_text_files_in_dir(project_folder, pattern=filename)
- if filename in matches:
- path = files.get_abs_path(project_folder, filename)
- return (files.normalize_a0_path(path), matches[filename])
+ for path, content in get_agents_md_chain(project_folder, project_folder):
+ return (files.normalize_a0_path(path), content)
return None
diff --git a/helpers/projects.py.dox.md b/helpers/projects.py.dox.md
index aae7514d0..e1e1b592e 100644
--- a/helpers/projects.py.dox.md
+++ b/helpers/projects.py.dox.md
@@ -47,6 +47,8 @@
- `reactivate_project_in_chats(name: str)`
- `deactivate_project_in_chats(name: str)`
- `build_system_prompt_vars(name: str)`
+- `get_agents_md_chain(root: str, target: str) -> list[tuple[str, str]]`
+- `build_agents_md_protocol(name: str, target: str | None=...) -> str`
- `get_additional_instructions_files(name: str)`
- `get_project_instruction_files(name: str, include_agents_md: bool=...) -> list[tuple[str, str]]`
- `get_project_agents_md_instruction_file(name: str) -> tuple[str, str] | None`
@@ -63,6 +65,8 @@
- Project extension data may add named top-level sections such as `llm`, but it must not overwrite core project fields owned by `EditProjectData`.
- Project extension save payloads exclude core project fields and transient inputs such as `git_token`; plugins needing core metadata should load it by project name.
- Project metadata setup creates and repairs `.a0proj/instructions`, `.a0proj/knowledge`, and `.a0proj/skills` so settings surfaces can open those folders consistently.
+- AGENTS.md discovery is a linear root-to-target chain walk with `AGENTS.override.md` precedence; sibling directories are not scanned.
+- Active-project AGENTS.md protocol guidance excludes the exact project root AGENTS.md because `build_system_prompt_vars(...)` already loads it into project instructions; prose for that protocol block lives in `prompts/agent.protocol.projects.agents_md.md`.
- Project MCP config uses the same JSON string shape as global MCP settings: an object with `mcpServers`.
- Project MCP load/save paths validate project names as simple folder basenames before touching `.a0proj/mcp_servers.json`.
- Observed side-effect areas: filesystem reads, filesystem writes, filesystem deletion, plugin state, settings/state persistence, secret handling.
diff --git a/prompts/agent.protocol.projects.agents_md.md b/prompts/agent.protocol.projects.agents_md.md
new file mode 100644
index 000000000..916f3c122
--- /dev/null
+++ b/prompts/agent.protocol.projects.agents_md.md
@@ -0,0 +1,6 @@
+# AGENTS.md instructions
+- these path-owned instructions are binding for the current project/workdir chain
+- active project root AGENTS.md is loaded separately in project instructions
+- before editing a deeper target, read AGENTS.md files on the direct path to that target; ignore siblings
+
+{{agents_md_instructions}}
diff --git a/tests/test_projects.py b/tests/test_projects.py
index 8dac0d2c1..0a2192ebb 100644
--- a/tests/test_projects.py
+++ b/tests/test_projects.py
@@ -107,6 +107,20 @@ def test_project_system_prompt_includes_root_agents_md_with_path(monkeypatch, tm
assert "Folder instruction rule." in instructions
+def test_project_system_prompt_prefers_agents_override_md(monkeypatch, tmp_path):
+ _prepare_project_tree(monkeypatch, tmp_path)
+ projects.create_project("demo", {"title": "Demo"})
+ project_root = tmp_path / "usr" / "projects" / "demo"
+ (project_root / "AGENTS.md").write_text("Standard rule.", encoding="utf-8")
+ (project_root / "AGENTS.override.md").write_text("Override rule.", encoding="utf-8")
+
+ instructions = projects.build_system_prompt_vars("demo")["project_instructions"]
+
+ assert "### path: /a0/usr/projects/demo/AGENTS.override.md" in instructions
+ assert "Override rule." in instructions
+ assert "Standard rule." not in instructions
+
+
def test_project_system_prompt_respects_disabled_agents_md(monkeypatch, tmp_path):
_prepare_project_tree(monkeypatch, tmp_path)
projects.create_project(
@@ -123,3 +137,51 @@ def test_project_system_prompt_respects_disabled_agents_md(monkeypatch, tmp_path
assert "Root AGENTS rule." not in prompt_vars["project_instructions"]
assert "AGENTS.md" not in prompt_vars["project_instructions"]
+
+
+def test_agents_md_chain_walks_direct_path_only(monkeypatch, tmp_path):
+ _prepare_project_tree(monkeypatch, tmp_path)
+ root = tmp_path
+ (root / "AGENTS.md").write_text("root doc", encoding="utf-8")
+ target = root / "services" / "payments"
+ sibling = root / "services" / "auth"
+ target.mkdir(parents=True)
+ sibling.mkdir(parents=True)
+ (root / "services" / "AGENTS.md").write_text("services doc", encoding="utf-8")
+ (target / "AGENTS.md").write_text("payments doc", encoding="utf-8")
+ (sibling / "AGENTS.md").write_text("auth doc", encoding="utf-8")
+
+ chain = projects.get_agents_md_chain(str(root), str(target / "handler.py"))
+ contents = [content for _, content in chain]
+
+ assert contents == ["root doc", "services doc", "payments doc"]
+
+
+def test_agents_md_protocol_excludes_project_root_and_keeps_subdir(
+ monkeypatch, tmp_path
+):
+ _prepare_project_tree(monkeypatch, tmp_path)
+ prompt_name = "agent.protocol.projects.agents_md.md"
+ prompt_source = Path(__file__).resolve().parents[1] / "prompts" / prompt_name
+ prompt_dir = tmp_path / "prompts"
+ prompt_dir.mkdir()
+ (prompt_dir / prompt_name).write_text(
+ prompt_source.read_text(encoding="utf-8"),
+ encoding="utf-8",
+ )
+ projects.create_project("demo", {"title": "Demo"})
+ (tmp_path / "AGENTS.md").write_text("framework doc", encoding="utf-8")
+ project_root = tmp_path / "usr" / "projects" / "demo"
+ (project_root / "AGENTS.md").write_text("project root doc", encoding="utf-8")
+ api_dir = project_root / "api"
+ api_dir.mkdir()
+ (api_dir / "AGENTS.md").write_text("api doc", encoding="utf-8")
+
+ protocol = projects.build_agents_md_protocol(
+ "demo",
+ target=str(api_dir / "handler.py"),
+ )
+
+ assert "framework doc" in protocol
+ assert "api doc" in protocol
+ assert "project root doc" not in protocol
diff --git a/tests/test_prompt_protocol.py b/tests/test_prompt_protocol.py
index c4d29bfdc..fcafcb7b5 100644
--- a/tests/test_prompt_protocol.py
+++ b/tests/test_prompt_protocol.py
@@ -94,6 +94,7 @@ async def test_project_prompt_moves_project_instructions_to_protocol(monkeypatch
"project_name": "Demo",
"project_description": "",
"project_instructions": "Project rule.",
+ "include_agents_md": True,
"project_path": "/a0/usr/projects/demo",
"project_git_url": "",
}
@@ -121,6 +122,11 @@ async def test_project_prompt_moves_project_instructions_to_protocol(monkeypatch
"build_system_prompt_vars",
lambda _name: project_vars,
)
+ monkeypatch.setattr(
+ project_prompt.projects,
+ "build_agents_md_protocol",
+ lambda _name: "AGENTS path rule.",
+ )
prompt = await project_prompt.build_prompt.__wrapped__( # type: ignore[attr-defined]
FakeAgent(),
@@ -128,4 +134,44 @@ async def test_project_prompt_moves_project_instructions_to_protocol(monkeypatch
)
assert "Project rule." not in prompt
+ assert "AGENTS path rule." not in prompt
+ assert list(loop_data.protocol_persistent) == [
+ "agents_md_instructions",
+ "project_instructions",
+ ]
+ assert "AGENTS path rule." in loop_data.protocol_persistent["agents_md_instructions"]
assert "Project rule." in loop_data.protocol_persistent["project_instructions"]
+
+
+@pytest.mark.asyncio
+async def test_project_prompt_does_not_load_agents_md_without_project(monkeypatch):
+ loop_data = LoopData()
+
+ class FakeContext:
+ def get_data(self, key):
+ assert key == project_prompt.projects.CONTEXT_DATA_KEY_PROJECT
+ return None
+
+ class FakeAgent:
+ context = FakeContext()
+
+ def read_prompt(self, prompt_file: str, **kwargs) -> str:
+ if prompt_file == "agent.system.projects.main.md":
+ return "project context may be active"
+ if prompt_file == "agent.system.projects.inactive.md":
+ return "no active project"
+ raise AssertionError(f"Unexpected prompt file: {prompt_file}")
+
+ monkeypatch.setattr(
+ project_prompt.projects,
+ "build_agents_md_protocol",
+ lambda _name: (_ for _ in ()).throw(AssertionError("unexpected AGENTS load")),
+ )
+
+ await project_prompt.build_prompt.__wrapped__( # type: ignore[attr-defined]
+ FakeAgent(),
+ loop_data=loop_data,
+ )
+
+ assert "agents_md_instructions" not in loop_data.protocol_persistent
+ assert "project_instructions" not in loop_data.protocol_persistent
From eb1326ccd327d7ad48048e4db787450154d077f5 Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Wed, 8 Jul 2026 14:38:43 +0200
Subject: [PATCH 28/43] Document Agent Zero runtime environment
Teach the agent prompt about the split framework and task Python runtimes so backend checks use /opt/venv-a0 while user-code tasks use /opt/venv.
Document the WebUI JSON API and CSRF token flow so terminal/API work can use the same session cookies and X-CSRF-Token correctly.
---
prompts/agent.system.main.environment.md | 12 +++++++++++-
1 file changed, 11 insertions(+), 1 deletion(-)
diff --git a/prompts/agent.system.main.environment.md b/prompts/agent.system.main.environment.md
index 53084d766..3dc08fefd 100644
--- a/prompts/agent.system.main.environment.md
+++ b/prompts/agent.system.main.environment.md
@@ -1,4 +1,14 @@
## Environment
live in kali linux docker container use debian kali packages
agent zero framework is python project in /a0 folder
-linux fully root accessible via terminal
\ No newline at end of file
+linux fully root accessible via terminal
+
+Python runtimes:
+- Framework runtime: /opt/venv-a0/bin/python runs Agent Zero itself, WebUI backend, API handlers, plugins/hooks, and framework imports.
+- Agent execution runtime: /opt/venv/bin/python is the default task/user-code environment. Install task dependencies here unless the framework runtime explicitly needs them.
+- Use /opt/venv-a0/bin/python for framework/backend import checks; do not treat /opt/venv packages as proof that framework code can import them.
+
+WebUI JSON API:
+- API handlers live at /api/<handler_name> and usually accept JSON POST requests.
+- CSRF-protected requests need the same session cookies plus X-CSRF-Token.
+- Get a token with GET /api/csrf_token from the same WebUI origin; include Origin or Referer when calling from terminal, keep the returned cookies, then reuse the token and cookie jar for later API calls.
From 6dfe33c4932dafe5e5f6039be1e20124d95951b8 Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Wed, 8 Jul 2026 15:08:44 +0200
Subject: [PATCH 29/43] Fix Codex OAuth request metadata
Send current Codex client metadata and compatibility headers on proxied Responses requests so ChatGPT OAuth calls pass upstream call checks.
Return Codex model slugs as the primary list while preserving model metadata for descriptions in the OAuth settings UI.
---
plugins/_oauth/AGENTS.md | 2 +
plugins/_oauth/api/models.py | 19 ++-
plugins/_oauth/helpers/codex.py | 153 +++++++++++++++++++--
plugins/_oauth/helpers/providers/codex.py | 5 +
plugins/_oauth/helpers/summary.py | 12 ++
plugins/_oauth/webui/config.html | 34 ++++-
plugins/_oauth/webui/oauth-config-store.js | 21 +++
tests/test_oauth_codex.py | 149 ++++++++++++++++++++
tests/test_oauth_providers.py | 41 ++++++
9 files changed, 420 insertions(+), 16 deletions(-)
diff --git a/plugins/_oauth/AGENTS.md b/plugins/_oauth/AGENTS.md
index efd6f98fa..7bf1f3348 100644
--- a/plugins/_oauth/AGENTS.md
+++ b/plugins/_oauth/AGENTS.md
@@ -30,6 +30,7 @@
- OAuth settings model slots must keep provider choice editable per slot, list only connected OAuth account providers, and persist the selected provider IDs into `chat_model.provider` and `utility_model.provider`.
- OAuth settings must dispatch `model-setup-changed` when a provider connection completes, but model selection remains explicit and must not be filled automatically.
- `helpers/providers/registry.py` is the source of truth for connectable OAuth providers.
+- The models API must preserve the legacy plain `models` slug list and may add `model_metadata` entries for richer provider catalogs.
- OAuth provider config must not expose the dummy `oauth` API key in `conf/model_providers.yaml`; the dummy key is a runtime-only shim supplied by the `get_api_key` extension after the account provider reports connected.
- Usage-plan metadata belongs only to connectable providers. Do not add metadata-only subscription families for providers this plugin cannot connect.
- API handlers should remain provider-aware. Missing or blank `provider_id` defaults to Codex only for existing backward compatibility; falsey non-string IDs must not silently default.
@@ -39,6 +40,7 @@
- Stored upstream base URLs and OAuth token endpoints must be validated against provider-owned allowlists before sending bearer or refresh tokens.
- Browser callback providers must support manual callback paste when the browser cannot reach the local callback route.
- Local proxy routes must remain loopback or token protected and must not add broad CORS access.
+- Codex Responses proxy requests must include Codex client metadata and compatibility headers such as `client_metadata`, `x-codex-installation-id`, `originator`, `session-id`, and `thread-id`.
## Work Guidance
diff --git a/plugins/_oauth/api/models.py b/plugins/_oauth/api/models.py
index ffc63afe3..493017ade 100644
--- a/plugins/_oauth/api/models.py
+++ b/plugins/_oauth/api/models.py
@@ -10,8 +10,23 @@ class Models(ApiHandler):
raw_provider_id = _provider_id(input)
try:
provider = get_provider(raw_provider_id)
- models = provider.models()
- return {"ok": True, "provider_id": provider.provider_id, "models": models}
+ model_catalog = getattr(provider, "model_catalog", None)
+ if callable(model_catalog):
+ catalog = model_catalog()
+ models = [
+ str(item.get("slug") or item.get("id") or "")
+ for item in catalog
+ if isinstance(item, dict) and (item.get("slug") or item.get("id"))
+ ]
+ else:
+ catalog = []
+ models = provider.models()
+ return {
+ "ok": True,
+ "provider_id": provider.provider_id,
+ "models": models,
+ "model_metadata": catalog,
+ }
except Exception as exc:
return {
"ok": False,
diff --git a/plugins/_oauth/helpers/codex.py b/plugins/_oauth/helpers/codex.py
index 824e19a92..a0fc20c5b 100644
--- a/plugins/_oauth/helpers/codex.py
+++ b/plugins/_oauth/helpers/codex.py
@@ -6,9 +6,11 @@ import hashlib
import json
import os
import secrets
+import stat
import subprocess
import threading
import time
+import uuid
from contextlib import contextmanager
from dataclasses import dataclass
from datetime import datetime, timedelta, timezone
@@ -33,9 +35,28 @@ except ImportError:
AUTH_FILENAME = "auth.json"
+INSTALLATION_ID_FILENAME = "installation_id"
ACCESS_EXPIRY_MARGIN = timedelta(minutes=5)
REFRESH_INTERVAL = timedelta(minutes=55)
DEFAULT_CODEX_MODEL = "gpt-5.5"
+CODEX_ORIGINATOR = "codex_cli_rs"
+CLIENT_METADATA_INSTALLATION_ID = "x-codex-installation-id"
+CLIENT_METADATA_WINDOW_ID = "x-codex-window-id"
+CLIENT_METADATA_KEYS = (
+ "slug",
+ "id",
+ "display_name",
+ "description",
+ "visibility",
+ "supported_in_api",
+ "default_reasoning_level",
+ "supported_reasoning_levels",
+ "additional_speed_tiers",
+ "service_tiers",
+ "context_window",
+ "max_context_window",
+ "priority",
+)
OAUTH_ERROR_KEYS = ("error_description", "error")
DEVICE_CODE_TIMEOUT_SECONDS = 15 * 60
WINDOWS_LOCK_RETRY_SECONDS = 0.05
@@ -565,11 +586,17 @@ def request_codex(
auth = load_auth()
target = build_upstream_url(path, cfg["upstream_base_url"])
request_headers = sanitize_forward_headers(headers or {})
+ metadata = client_metadata_from_body(body) or build_client_metadata()
request_headers.update(
{
"Authorization": f"Bearer {auth.access_token}",
"chatgpt-account-id": auth.account_id,
"OpenAI-Beta": "responses=experimental",
+ "originator": CODEX_ORIGINATOR,
+ CLIENT_METADATA_INSTALLATION_ID: metadata[CLIENT_METADATA_INSTALLATION_ID],
+ CLIENT_METADATA_WINDOW_ID: metadata[CLIENT_METADATA_WINDOW_ID],
+ "session-id": metadata["session_id"],
+ "thread-id": metadata["thread_id"],
}
)
@@ -584,11 +611,14 @@ def request_codex(
)
-def fetch_models() -> list[str]:
+def fetch_model_catalog() -> list[dict[str, Any]]:
cfg = codex_config()
configured = cfg["models"]
if configured:
- return configured
+ return [
+ {"slug": model, "id": model, "display_name": model}
+ for model in configured
+ ]
client_version = resolve_codex_version()
params = {"client_version": client_version} if client_version else None
@@ -601,31 +631,103 @@ def fetch_models() -> list[str]:
payload = response.json()
raw_models = payload.get("models") if isinstance(payload, dict) else None
+ if raw_models is None and isinstance(payload, dict):
+ raw_models = payload.get("data")
if not isinstance(raw_models, list):
raise RuntimeError("Codex returned a malformed models response.")
- models: list[str] = []
+ catalog: list[dict[str, Any]] = []
seen: set[str] = set()
for item in raw_models:
- slug = item.get("slug") if isinstance(item, dict) else None
- if isinstance(slug, str) and slug and slug not in seen:
+ if isinstance(item, dict):
+ slug = _string(item.get("slug") or item.get("id"))
+ model = {
+ key: item[key]
+ for key in CLIENT_METADATA_KEYS
+ if key in item
+ }
+ else:
+ slug = _string(item)
+ model = {}
+ if slug and slug not in seen:
seen.add(slug)
- models.append(slug)
- if not models:
+ model["slug"] = slug
+ model.setdefault("id", slug)
+ model.setdefault("display_name", slug)
+ catalog.append(model)
+ if not catalog:
raise RuntimeError("Codex returned an empty models list.")
- return models
+ return catalog
+
+
+def fetch_models() -> list[str]:
+ return [model["slug"] for model in fetch_model_catalog()]
def prepare_responses_body(body: dict[str, Any], *, force_stream: bool) -> dict[str, Any]:
normalized = dict(body)
normalized.setdefault("instructions", "")
normalized.setdefault("store", False)
+ normalized["client_metadata"] = merge_client_metadata(normalized.get("client_metadata"))
if force_stream:
normalized["stream"] = True
+ if isinstance(normalized.get("reasoning"), dict):
+ include = normalized.get("include")
+ values = list(include) if isinstance(include, list) else []
+ if "reasoning.encrypted_content" not in values:
+ values.append("reasoning.encrypted_content")
+ normalized["include"] = values
normalized.pop("max_output_tokens", None)
return normalized
+def build_client_metadata() -> dict[str, str]:
+ request_id = f"agent-zero-{uuid.uuid4()}"
+ return {
+ CLIENT_METADATA_INSTALLATION_ID: resolve_installation_id(),
+ "session_id": request_id,
+ "thread_id": request_id,
+ CLIENT_METADATA_WINDOW_ID: "agent-zero",
+ }
+
+
+def merge_client_metadata(value: Any) -> dict[str, str]:
+ metadata = {
+ str(key): str(item)
+ for key, item in (value.items() if isinstance(value, dict) else [])
+ if item is not None and str(item)
+ }
+ metadata.update(build_client_metadata())
+ return metadata
+
+
+def client_metadata_from_body(body: bytes | str | None) -> dict[str, str] | None:
+ if body is None:
+ return None
+ try:
+ text = body.decode("utf-8") if isinstance(body, bytes) else body
+ payload = json.loads(text)
+ except Exception:
+ return None
+ if not isinstance(payload, dict):
+ return None
+ metadata = payload.get("client_metadata")
+ if not isinstance(metadata, dict):
+ return None
+ result = {
+ str(key): str(value)
+ for key, value in metadata.items()
+ if value is not None and str(value)
+ }
+ required = {
+ CLIENT_METADATA_INSTALLATION_ID,
+ "session_id",
+ "thread_id",
+ CLIENT_METADATA_WINDOW_ID,
+ }
+ return result if required.issubset(result) else None
+
+
def collect_completed_response(response: requests.Response) -> dict[str, Any]:
latest_response: dict[str, Any] | None = None
latest_error: Any = None
@@ -728,9 +830,7 @@ def extract_sse_text_deltas(payload: dict[str, Any], event_type: str = "") -> li
if (payload.get("type") or event_type) in {
"response.output_text.delta",
- "response.output_text.done",
"response.text.delta",
- "response.text.done",
}:
_append_text_value(pieces, payload.get("text"))
@@ -985,6 +1085,39 @@ def resolve_codex_version() -> str:
return ""
+def resolve_installation_id() -> str:
+ for path in installation_id_candidates():
+ try:
+ value = path.read_text(encoding="utf-8").strip()
+ except OSError:
+ continue
+ if value:
+ return value
+
+ value = str(uuid.uuid4())
+ path = plugin_installation_id_path()
+ try:
+ path.parent.mkdir(parents=True, exist_ok=True)
+ path.write_text(value, encoding="utf-8")
+ os.chmod(path, stat.S_IRUSR | stat.S_IWUSR)
+ except OSError:
+ pass
+ return value
+
+
+def installation_id_candidates() -> list[Path]:
+ candidates = [plugin_installation_id_path()]
+ codex_home = os.environ.get("CODEX_HOME")
+ if codex_home:
+ candidates.append(Path(codex_home).expanduser() / INSTALLATION_ID_FILENAME)
+ candidates.append(Path.home() / ".codex" / INSTALLATION_ID_FILENAME)
+ return candidates
+
+
+def plugin_installation_id_path() -> Path:
+ return Path(files.get_abs_path("usr", "plugins", "_oauth", "codex", INSTALLATION_ID_FILENAME))
+
+
def resolve_auth_file_candidates() -> list[Path]:
return [resolve_auth_write_path()]
diff --git a/plugins/_oauth/helpers/providers/codex.py b/plugins/_oauth/helpers/providers/codex.py
index d619927da..0a59a79a0 100644
--- a/plugins/_oauth/helpers/providers/codex.py
+++ b/plugins/_oauth/helpers/providers/codex.py
@@ -232,6 +232,11 @@ class CodexOAuthProvider:
return codex.fetch_models()
+ def model_catalog(self) -> list[dict[str, Any]]:
+ from plugins._oauth.helpers import codex
+
+ return codex.fetch_model_catalog()
+
def disconnect(self) -> dict[str, Any]:
from plugins._oauth.helpers import codex
diff --git a/plugins/_oauth/helpers/summary.py b/plugins/_oauth/helpers/summary.py
index 107228811..2585859e1 100644
--- a/plugins/_oauth/helpers/summary.py
+++ b/plugins/_oauth/helpers/summary.py
@@ -63,13 +63,25 @@ def _provider_status(provider: Any) -> dict[str, Any]:
status["usage_windows"] = usage_windows
return status
except Exception as exc:
+ metadata = _provider_metadata(provider)
return {
+ **metadata,
"provider_id": str(getattr(provider, "provider_id", "")),
"connected": False,
"error": str(exc),
}
+def _provider_metadata(provider: Any) -> dict[str, Any]:
+ try:
+ metadata = provider.metadata()
+ to_dict = getattr(metadata, "to_dict", None)
+ value = to_dict() if callable(to_dict) else metadata
+ return value if isinstance(value, dict) else {}
+ except Exception:
+ return {}
+
+
def _account_summary(provider: dict[str, Any], catalog: dict[str, Any]) -> dict[str, Any]:
provider_id = str(provider.get("provider_id") or "")
plan_entry = catalog.get(provider_id) if isinstance(catalog, dict) else None
diff --git a/plugins/_oauth/webui/config.html b/plugins/_oauth/webui/config.html
index 72177d99d..2e04cd876 100644
--- a/plugins/_oauth/webui/config.html
+++ b/plugins/_oauth/webui/config.html
@@ -259,8 +259,16 @@
type="button"
class="oauth-model-item"
@click="$store.oauthConfig.selectModel(slot.key, model)"
- x-text="model"
- ></button>
+ >
+ <span
+ class="oauth-model-name"
+ x-text="model"
+ ></span>
+ <small
+ x-show="$store.oauthConfig.modelDescription(slot.provider, model)"
+ x-text="$store.oauthConfig.modelDescription(slot.provider, model)"
+ ></small>
+ </button>
</template>
<div class="oauth-model-item muted" x-show="$store.oauthConfig.filteredModels(slot.key).length === 0">
No models found. You can still type the model name manually.
@@ -282,7 +290,10 @@
</div>
<div class="oauth-models">
<template x-for="model in $store.oauthConfig.activeProviderModels()" :key="model">
- <span x-text="model"></span>
+ <span
+ :title="$store.oauthConfig.modelDescription($store.oauthConfig.activeModelProvider, model)"
+ x-text="model"
+ ></span>
</template>
</div>
</section>
@@ -991,7 +1002,9 @@
}
.oauth-model-item {
- display: block;
+ display: flex;
+ flex-direction: column;
+ gap: 3px;
width: 100%;
min-height: 34px;
padding: 8px 10px;
@@ -1004,6 +1017,19 @@
cursor: pointer;
}
+ .oauth-model-name {
+ font-weight: 650;
+ }
+
+ .oauth-model-item small {
+ overflow: hidden;
+ color: var(--color-text-secondary);
+ font-size: 0.72rem;
+ line-height: 1.3;
+ text-overflow: ellipsis;
+ white-space: nowrap;
+ }
+
.oauth-model-item:hover,
.oauth-model-item:focus-visible {
background: color-mix(in srgb, var(--color-border) 42%, transparent);
diff --git a/plugins/_oauth/webui/oauth-config-store.js b/plugins/_oauth/webui/oauth-config-store.js
index c77701631..004093662 100644
--- a/plugins/_oauth/webui/oauth-config-store.js
+++ b/plugins/_oauth/webui/oauth-config-store.js
@@ -107,6 +107,7 @@ export const store = createStore("oauthConfig", {
disconnecting: false,
loadingModels: false,
providerModels: {},
+ providerModelMetadata: {},
providerUi: providerUiDefaults(),
connectingProvider: "",
disconnectingProvider: "",
@@ -150,6 +151,7 @@ export const store = createStore("oauthConfig", {
this.disconnecting = false;
this.loadingModels = false;
this.providerModels = {};
+ this.providerModelMetadata = {};
this.providerUi = providerUiDefaults();
this.connectingProvider = "";
this.disconnectingProvider = "";
@@ -556,6 +558,14 @@ export const store = createStore("oauthConfig", {
return this.providerModels[this.activeModelProvider] || [];
},
+ modelMetadata(providerId, model) {
+ return this.providerModelMetadata[providerId]?.[model] || {};
+ },
+
+ modelDescription(providerId, model) {
+ return this.modelMetadata(providerId, model).description || "";
+ },
+
activeModelsDescription() {
if (!this.providerConnected(this.activeModelProvider)) return "";
return `Available models from ${this.providerLabel(this.activeModelProvider)}`;
@@ -967,12 +977,20 @@ export const store = createStore("oauthConfig", {
throw new Error(response?.error || `Could not load ${this.providerLabel(selectedProvider)} models.`);
}
const models = Array.isArray(response.models) ? response.models : [];
+ const metadata = Array.isArray(response.model_metadata) ? response.model_metadata : [];
+ const metadataMap = metadata.reduce((result, item) => {
+ const key = String(item?.slug || item?.id || "");
+ if (key) result[key] = item;
+ return result;
+ }, {});
this.providerModels = { ...this.providerModels, [selectedProvider]: models };
+ this.providerModelMetadata = { ...this.providerModelMetadata, [selectedProvider]: metadataMap };
this.models = models;
if (openDropdown) this.openModelDropdown(openDropdown);
if (!silent) void toastFrontendSuccess(`${this.providerLabel(selectedProvider)} models loaded.`, "OAuth Connections");
} catch (error) {
this.providerModels = { ...this.providerModels, [selectedProvider]: [] };
+ this.providerModelMetadata = { ...this.providerModelMetadata, [selectedProvider]: {} };
this.models = [];
if (!silent) void toastFrontendError(messageOf(error), "OAuth Connections");
} finally {
@@ -999,6 +1017,9 @@ export const store = createStore("oauthConfig", {
const providerModels = { ...this.providerModels };
delete providerModels[providerId];
this.providerModels = providerModels;
+ const providerModelMetadata = { ...this.providerModelMetadata };
+ delete providerModelMetadata[providerId];
+ this.providerModelMetadata = providerModelMetadata;
if (this.activeModelProvider === providerId) this.models = [];
this.clearProviderDevice(providerId);
if (this.connectingProvider === providerId) this.connectingProvider = "";
diff --git a/tests/test_oauth_codex.py b/tests/test_oauth_codex.py
index 89c0a6e17..396584c18 100644
--- a/tests/test_oauth_codex.py
+++ b/tests/test_oauth_codex.py
@@ -137,6 +137,145 @@ def test_codex_fetch_models_omits_fake_client_version(monkeypatch):
assert calls == [("/models", None)]
+def test_codex_fetch_model_catalog_preserves_model_metadata(monkeypatch):
+ class FakeResponse:
+ ok = True
+
+ def json(self):
+ return {
+ "models": [
+ {
+ "slug": "gpt-5.5",
+ "display_name": "GPT-5.5",
+ "description": "Frontier coding model.",
+ "default_reasoning_level": "medium",
+ "base_instructions": "too large for the settings UI",
+ }
+ ]
+ }
+
+ monkeypatch.setattr(codex, "codex_config", lambda: {"models": []})
+ monkeypatch.setattr(codex, "resolve_codex_version", lambda: "0.142.5")
+ monkeypatch.setattr(codex, "request_codex", lambda path, params=None: FakeResponse())
+
+ assert codex.fetch_model_catalog() == [
+ {
+ "slug": "gpt-5.5",
+ "id": "gpt-5.5",
+ "display_name": "GPT-5.5",
+ "description": "Frontier coding model.",
+ "default_reasoning_level": "medium",
+ }
+ ]
+
+
+def test_prepare_responses_body_adds_codex_client_metadata(monkeypatch):
+ monkeypatch.setattr(
+ codex,
+ "build_client_metadata",
+ lambda: {
+ "x-codex-installation-id": "install-1",
+ "session_id": "session-1",
+ "thread_id": "thread-1",
+ "x-codex-window-id": "agent-zero",
+ },
+ )
+
+ body = codex.prepare_responses_body(
+ {
+ "model": "gpt-5.5",
+ "input": "hello",
+ "client_metadata": {
+ "caller": "plugin-test",
+ "x-codex-installation-id": "stale",
+ },
+ "reasoning": {"effort": "medium"},
+ "include": ["output_text"],
+ },
+ force_stream=True,
+ )
+
+ assert body["client_metadata"] == {
+ "caller": "plugin-test",
+ "x-codex-installation-id": "install-1",
+ "session_id": "session-1",
+ "thread_id": "thread-1",
+ "x-codex-window-id": "agent-zero",
+ }
+ assert body["stream"] is True
+ assert body["include"] == ["output_text", "reasoning.encrypted_content"]
+
+
+def test_request_codex_sends_current_codex_headers_from_body(monkeypatch):
+ calls = []
+ body = json.dumps(
+ {
+ "client_metadata": {
+ "x-codex-installation-id": "install-1",
+ "session_id": "session-1",
+ "thread_id": "thread-1",
+ "x-codex-window-id": "agent-zero",
+ }
+ }
+ )
+
+ class FakeResponse:
+ ok = True
+
+ monkeypatch.setattr(
+ codex,
+ "codex_config",
+ lambda: {
+ "upstream_base_url": "https://chatgpt.example/backend-api/codex",
+ "request_timeout_seconds": 120,
+ },
+ )
+ monkeypatch.setattr(
+ codex,
+ "load_auth",
+ lambda: codex.EffectiveAuth(
+ access_token="access-token",
+ account_id="account-1",
+ ),
+ )
+
+ def fake_request(method, target, headers, data, params, timeout, stream):
+ calls.append(
+ {
+ "method": method,
+ "target": target,
+ "headers": headers,
+ "data": data,
+ "params": params,
+ "timeout": timeout,
+ "stream": stream,
+ }
+ )
+ return FakeResponse()
+
+ monkeypatch.setattr(codex.requests, "request", fake_request)
+
+ response = codex.request_codex(
+ "/responses",
+ method="POST",
+ headers={"Content-Type": "application/json"},
+ body=body,
+ stream=True,
+ )
+
+ assert response.ok is True
+ call = calls[0]
+ assert call["target"] == "https://chatgpt.example/backend-api/codex/responses"
+ assert call["headers"]["Authorization"] == "Bearer access-token"
+ assert call["headers"]["chatgpt-account-id"] == "account-1"
+ assert call["headers"]["OpenAI-Beta"] == "responses=experimental"
+ assert call["headers"]["originator"] == "codex_cli_rs"
+ assert call["headers"]["x-codex-installation-id"] == "install-1"
+ assert call["headers"]["session-id"] == "session-1"
+ assert call["headers"]["thread-id"] == "thread-1"
+ assert call["headers"]["x-codex-window-id"] == "agent-zero"
+
+
def test_chat_messages_to_response_body_preserves_image_parts_for_responses():
data_url = "data:image/png;base64,abcd"
@@ -231,6 +370,16 @@ def test_extract_sse_text_deltas_reads_chat_completion_chunks():
assert deltas == ["Hel", "lo"]
+def test_extract_sse_text_deltas_ignores_final_done_text():
+ assert (
+ codex.extract_sse_text_deltas(
+ {"type": "response.output_text.done", "text": "Hello"},
+ "response.output_text.done",
+ )
+ == []
+ )
+
+
def test_collect_completed_response_falls_back_to_text_deltas():
class FakeResponse:
encoding = "utf-8"
diff --git a/tests/test_oauth_providers.py b/tests/test_oauth_providers.py
index 155c2bcba..febccead0 100644
--- a/tests/test_oauth_providers.py
+++ b/tests/test_oauth_providers.py
@@ -390,6 +390,13 @@ def test_status_api_contains_provider_status_exceptions(monkeypatch):
class FailingProvider:
provider_id = XAI_GROK_PROVIDER_ID
+ def metadata(self):
+ return {
+ "provider_id": XAI_GROK_PROVIDER_ID,
+ "display_name": "xAI Grok",
+ "auth_flow": "browser_pkce",
+ }
+
def status(self):
raise RuntimeError("status failed")
@@ -409,6 +416,8 @@ def test_status_api_contains_provider_status_exceptions(monkeypatch):
assert response["provider_map"][CODEX_PROVIDER_ID]["connected"] is True
assert response["provider_map"][XAI_GROK_PROVIDER_ID] == {
"provider_id": XAI_GROK_PROVIDER_ID,
+ "display_name": "xAI Grok",
+ "auth_flow": "browser_pkce",
"connected": False,
"error": "status failed",
}
@@ -951,6 +960,38 @@ def test_manual_callback_unknown_provider_returns_structured_error():
assert "Unknown OAuth provider" in response["error"]
+def test_models_api_returns_optional_model_metadata(monkeypatch):
+ class FakeProvider:
+ provider_id = CODEX_PROVIDER_ID
+
+ def model_catalog(self):
+ return [
+ {
+ "slug": "gpt-5.5",
+ "display_name": "GPT-5.5",
+ "description": "Frontier coding model.",
+ }
+ ]
+
+ def models(self):
+ return ["ignored-when-catalog-exists"]
+
+ models_module = sys.modules["plugins._oauth.api.models"]
+ monkeypatch.setattr(models_module, "get_provider", lambda provider_id: FakeProvider())
+
+ response = asyncio.run(Models(None, None).process({"provider_id": CODEX_PROVIDER_ID}, FakeRequest()))
+
+ assert response["ok"] is True
+ assert response["models"] == ["gpt-5.5"]
+ assert response["model_metadata"] == [
+ {
+ "slug": "gpt-5.5",
+ "display_name": "GPT-5.5",
+ "description": "Frontier coding model.",
+ }
+ ]
+
+
def test_unknown_provider_id_on_provider_aware_api_returns_structured_error():
response = asyncio.run(Models(None, None).process({"provider_id": "missing"}, FakeRequest()))
From 4f80ded5318158b42ac18d2214a3faa9778dda55 Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Wed, 8 Jul 2026 15:37:18 +0200
Subject: [PATCH 30/43] Fix transient UI layering over modals
Raise notification toasts and Bootstrap tooltips above normal and legacy modal overlays while keeping confirmation dialogs on top.
Remove the duplicate tooltip CSS block and add a focused regression for the layer ordering contract.
---
tests/test_browser_agent_regressions.py | 23 ++++
webui/AGENTS.md | 1 +
webui/components/notifications/AGENTS.md | 1 +
.../notification-toast-stack.html | 2 +-
webui/index.css | 110 +-----------------
5 files changed, 28 insertions(+), 109 deletions(-)
diff --git a/tests/test_browser_agent_regressions.py b/tests/test_browser_agent_regressions.py
index 2a42f4b55..ce921faa9 100644
--- a/tests/test_browser_agent_regressions.py
+++ b/tests/test_browser_agent_regressions.py
@@ -969,6 +969,29 @@ def test_surface_buttons_keep_modal_and_canvas_entry_points_separate():
assert "/plugins/_desktop/webui/main.html" in surfaces_js
+def test_transient_ui_layers_above_modals_below_confirm_dialogs():
+ index_css = (PROJECT_ROOT / "webui" / "index.css").read_text(encoding="utf-8")
+ modals_js = (PROJECT_ROOT / "webui" / "js" / "modals.js").read_text(encoding="utf-8")
+ modals_css = (PROJECT_ROOT / "webui" / "css" / "modals.css").read_text(encoding="utf-8")
+ toast_html = (
+ PROJECT_ROOT / "webui" / "components" / "notifications" / "notification-toast-stack.html"
+ ).read_text(encoding="utf-8")
+
+ modal_base_z = int(re.search(r"const baseZIndex = (\d+);", modals_js).group(1))
+ legacy_overlay_z = int(re.search(r"\.modal-overlay \{[^}]*z-index: (\d+);", modals_css, re.S).group(1))
+ confirm_z = int(re.search(r"\.confirm-dialog-backdrop \{[^}]*z-index: (\d+);", modals_css, re.S).group(1))
+ toast_z = int(re.search(r"\.toast-stack-container \{[^}]*z-index: (\d+);", toast_html, re.S).group(1))
+ tooltip_z_indexes = [
+ int(z)
+ for z in re.findall(r"\.tooltip \{[^}]*z-index: (\d+);", index_css, re.S)
+ ]
+
+ assert tooltip_z_indexes
+ assert toast_z > max(modal_base_z + 20, legacy_overlay_z)
+ assert all(z > toast_z for z in tooltip_z_indexes)
+ assert all(z < confirm_z for z in [toast_z, *tooltip_z_indexes])
+
+
def test_browser_tool_does_not_auto_open_canvas_policy_is_documented():
prompt = (
PROJECT_ROOT / "plugins" / "_browser" / "prompts" / "agent.system.tool.browser.md"
diff --git a/webui/AGENTS.md b/webui/AGENTS.md
index 628a3a17b..63c4fa40e 100644
--- a/webui/AGENTS.md
+++ b/webui/AGENTS.md
@@ -28,6 +28,7 @@
- Startup scripts in `index.html` must be local and non-parser-blocking: use ES modules, `defer`, or `async` for every script with `src`.
- Rubik (`--font-family-main`) is the default WebUI text and control font; use the code/mono font tokens only for code, logs, paths, and fixed-width data.
- Hover, focus, and active border treatments should follow existing neutral border/background patterns; avoid hard-coded blue border highlights unless matching an established specialized surface.
+- Keep transient UI affordances such as Bootstrap tooltips and notification toasts above normal and legacy modal layers while confirmation dialogs remain on top.
## Work Guidance
diff --git a/webui/components/notifications/AGENTS.md b/webui/components/notifications/AGENTS.md
index 071bf076f..4e958d3c1 100644
--- a/webui/components/notifications/AGENTS.md
+++ b/webui/components/notifications/AGENTS.md
@@ -15,6 +15,7 @@
- Use the notification system for user-facing success, warning, info, and error feedback.
- Keep public helper names stable for core and plugin callers.
+- Keep the toast stack visible above normal and legacy modal overlays while staying below confirmation dialogs.
- Avoid exposing secrets or raw auth payloads in notification text.
## Work Guidance
diff --git a/webui/components/notifications/notification-toast-stack.html b/webui/components/notifications/notification-toast-stack.html
index 502dabdb9..3a9fc29c1 100644
--- a/webui/components/notifications/notification-toast-stack.html
+++ b/webui/components/notifications/notification-toast-stack.html
@@ -58,7 +58,7 @@
position: absolute;
bottom: 5px; /* Spacing from the bottom of the zero-height container */
padding-right: 5px;
- z-index: 4000;
+ z-index: 8900;
display: flex;
flex-direction: column-reverse; /* Stack toasts upwards */
gap: 8px;
diff --git a/webui/index.css b/webui/index.css
index 3a408a7ef..5b1362c6f 100644
--- a/webui/index.css
+++ b/webui/index.css
@@ -189,114 +189,8 @@ body,
/* Bootstrap-like tooltips */
.tooltip {
position: absolute;
- /* Above custom modals (base z-index: 3000 in `webui/js/modals.js`) */
- z-index: 5000;
- display: block;
- margin: 0;
- font-family: var(--font-family-main);
- font-style: normal;
- font-weight: 400;
- line-height: 1.2;
- text-align: left;
- text-decoration: none;
- text-shadow: none;
- text-transform: none;
- letter-spacing: normal;
- word-break: normal;
- white-space: normal;
- opacity: 0;
- pointer-events: none;
- transition: opacity 0.08s ease-in-out;
-}
-
-.tooltip.show {
- opacity: 1;
-}
-
-.tooltip .tooltip-inner {
- max-width: 240px;
- padding: 0.35rem 0.6rem;
- color: var(--color-text);
- background-color: var(--color-secondary);
- border-radius: 0.375rem;
- box-shadow: 0 4px 12px rgba(0, 0, 0, 0.25);
- font-size: 0.75rem;
-}
-
-.tooltip .tooltip-arrow {
- position: absolute;
- width: 0.8rem;
- height: 0.4rem;
-}
-
-.tooltip .tooltip-arrow::before {
- content: "";
- position: absolute;
- border-color: transparent;
- border-style: solid;
-}
-
-.bs-tooltip-top .tooltip-arrow {
- bottom: 0;
- left: 50%;
- transform: translateX(-50%);
-}
-
-.bs-tooltip-top .tooltip-arrow::before {
- border-width: 0.4rem 0.4rem 0;
- border-top-color: var(--color-secondary);
- top: 0;
-}
-
-.bs-tooltip-bottom .tooltip-arrow {
- top: 0;
- left: 50%;
- transform: translateX(-50%);
-}
-
-.bs-tooltip-bottom .tooltip-arrow::before {
- border-width: 0 0.4rem 0.4rem;
- border-bottom-color: var(--color-secondary);
- top: 0;
-}
-
-.bs-tooltip-start .tooltip-arrow,
-.bs-tooltip-left .tooltip-arrow {
- right: 0;
- top: 50%;
- width: 0.4rem;
- height: 0.8rem;
- transform: translateY(-50%);
-}
-
-.bs-tooltip-start .tooltip-arrow::before,
-.bs-tooltip-left .tooltip-arrow::before {
- border-width: 0.4rem 0 0.4rem 0.4rem;
- border-left-color: var(--color-secondary);
- left: 0;
-}
-
-.bs-tooltip-end .tooltip-arrow,
-.bs-tooltip-right .tooltip-arrow {
- left: 0;
- top: 50%;
- width: 0.4rem;
- height: 0.8rem;
- transform: translateY(-50%);
-}
-
-.bs-tooltip-end .tooltip-arrow::before,
-.bs-tooltip-right .tooltip-arrow::before {
- border-width: 0.4rem 0.4rem 0.4rem 0;
- border-right-color: var(--color-secondary);
- right: 0;
-}
-
-/* Bootstrap-like tooltips */
-.tooltip {
- position: absolute;
- /* Above custom modals (base z-index: 3000 in `webui/js/modals.js`) */
- z-index: 5000;
+ /* Above custom modals and toasts, below confirm dialogs. */
+ z-index: 9000;
display: block;
margin: 0;
font-family: var(--font-family-main);
From a7255d97734abe511f04f6ffa4b6aa076270d2bd Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Wed, 8 Jul 2026 20:53:56 +0200
Subject: [PATCH 31/43] Await cached component module imports
Ensure x-component loading awaits cached dynamic import promises before appending deferred markup, matching first-load behavior and preventing Alpine bindings from running before imported stores register.
Document the loader contract in the WebUI JS DOX notes.
---
webui/js/AGENTS.md | 1 +
webui/js/components.js | 4 ++--
2 files changed, 3 insertions(+), 2 deletions(-)
diff --git a/webui/js/AGENTS.md b/webui/js/AGENTS.md
index e92975fb5..60497572f 100644
--- a/webui/js/AGENTS.md
+++ b/webui/js/AGENTS.md
@@ -37,6 +37,7 @@
- Keep extension loader cache keys and extension point names stable for plugins.
- HTML extension loading turns discovered HTML files into `<x-component>` tags; JavaScript extensions must export a default function.
- `<x-component>` loading must process component `style`, `script`, and stylesheet-link assets only once, even when a component keeps its scoped `<style>` inside `<body>`.
+- Every `<x-component>` instance must await cached module-load promises before markup is appended so Alpine bindings only run after imported stores exist.
- Frontend extension hooks such as `confirm_dialog_after_render` and `get_tool_message_handler` must preserve their mutable context contracts.
- Sanitize or safely render user/model-provided HTML and markdown.
- Do not expose secrets in localStorage, console logs, URLs, or WebSocket payloads.
diff --git a/webui/js/components.js b/webui/js/components.js
index 3e6f9895b..ac8144604 100644
--- a/webui/js/components.js
+++ b/webui/js/components.js
@@ -79,8 +79,8 @@ export async function importComponent(path, targetElement) {
if (!componentCache[resolvedUrl]) {
const modulePromise = import(resolvedUrl);
componentCache[resolvedUrl] = modulePromise;
- loadPromises.push(modulePromise);
}
+ loadPromises.push(componentCache[resolvedUrl]);
} else {
const virtualUrl = `${componentUrl.replaceAll(
"/",
@@ -122,8 +122,8 @@ export async function importComponent(path, targetElement) {
.finally(() => URL.revokeObjectURL(blobUrl));
componentCache[virtualUrl] = modulePromise;
- loadPromises.push(modulePromise);
}
+ loadPromises.push(componentCache[virtualUrl]);
}
} else {
// Non-module script
From 5f144350936c58f07cb1aef8d0b5b83babe94786 Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Thu, 9 Jul 2026 02:41:03 +0200
Subject: [PATCH 32/43] Normalize Codex Responses input shape
Ensure the Codex OAuth Responses proxy forwards list-shaped input for empty continuation turns while preserving string prompts as user input items. Add regression coverage for the gpt-5.5 continuation case and document the proxy contract.
---
plugins/_oauth/AGENTS.md | 2 +-
plugins/_oauth/helpers/codex.py | 7 +++++++
tests/test_oauth_codex.py | 17 +++++++++++++++++
3 files changed, 25 insertions(+), 1 deletion(-)
diff --git a/plugins/_oauth/AGENTS.md b/plugins/_oauth/AGENTS.md
index 7bf1f3348..ebcf18c36 100644
--- a/plugins/_oauth/AGENTS.md
+++ b/plugins/_oauth/AGENTS.md
@@ -40,7 +40,7 @@
- Stored upstream base URLs and OAuth token endpoints must be validated against provider-owned allowlists before sending bearer or refresh tokens.
- Browser callback providers must support manual callback paste when the browser cannot reach the local callback route.
- Local proxy routes must remain loopback or token protected and must not add broad CORS access.
-- Codex Responses proxy requests must include Codex client metadata and compatibility headers such as `client_metadata`, `x-codex-installation-id`, `originator`, `session-id`, and `thread-id`.
+- Codex Responses proxy requests must include Codex client metadata and compatibility headers such as `client_metadata`, `x-codex-installation-id`, `originator`, `session-id`, and `thread-id`, and must forward `input` as a list for upstream Codex compatibility.
## Work Guidance
diff --git a/plugins/_oauth/helpers/codex.py b/plugins/_oauth/helpers/codex.py
index a0fc20c5b..adcfdaec6 100644
--- a/plugins/_oauth/helpers/codex.py
+++ b/plugins/_oauth/helpers/codex.py
@@ -666,6 +666,13 @@ def fetch_models() -> list[str]:
def prepare_responses_body(body: dict[str, Any], *, force_stream: bool) -> dict[str, Any]:
normalized = dict(body)
+ input_value = normalized.get("input")
+ if isinstance(input_value, str):
+ normalized["input"] = (
+ [{"role": "user", "content": input_value}] if input_value else []
+ )
+ elif not isinstance(input_value, list):
+ normalized["input"] = []
normalized.setdefault("instructions", "")
normalized.setdefault("store", False)
normalized["client_metadata"] = merge_client_metadata(normalized.get("client_metadata"))
diff --git a/tests/test_oauth_codex.py b/tests/test_oauth_codex.py
index 396584c18..fc4a3159f 100644
--- a/tests/test_oauth_codex.py
+++ b/tests/test_oauth_codex.py
@@ -202,10 +202,27 @@ def test_prepare_responses_body_adds_codex_client_metadata(monkeypatch):
"thread_id": "thread-1",
"x-codex-window-id": "agent-zero",
}
+ assert body["input"] == [{"role": "user", "content": "hello"}]
assert body["stream"] is True
assert body["include"] == ["output_text", "reasoning.encrypted_content"]
+def test_prepare_responses_body_sends_empty_continuation_input_as_list(monkeypatch):
+ monkeypatch.setattr(codex, "build_client_metadata", lambda: {})
+
+ body = codex.prepare_responses_body(
+ {
+ "model": "gpt-5.5",
+ "input": "",
+ "previous_response_id": "resp_1",
+ },
+ force_stream=True,
+ )
+
+ assert body["input"] == []
+ assert body["previous_response_id"] == "resp_1"
+
+
def test_request_codex_sends_current_codex_headers_from_body(monkeypatch):
calls = []
body = json.dumps(
From f1787b65f5914f611d353a3d6b47d43baecfd032 Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Thu, 9 Jul 2026 03:29:52 +0200
Subject: [PATCH 33/43] Fix code execution PTY reset hangs
Start local PTY shells in a new process session and make close() escalate from SIGTERM to SIGKILL when a foreground command refuses to exit. This keeps code_execution_tool reset from blocking indefinitely on stuck terminal processes.\n\nAdd regression coverage for closing a TERM-resistant shell and document the reset lifecycle contract.
---
plugins/_code_execution/AGENTS.md | 1 +
.../_code_execution/helpers/tty_session.py | 50 +++++++++++++------
tests/test_code_execution_pager.py | 15 +++++-
3 files changed, 51 insertions(+), 15 deletions(-)
diff --git a/plugins/_code_execution/AGENTS.md b/plugins/_code_execution/AGENTS.md
index 2de8457f1..1f79ed4ee 100644
--- a/plugins/_code_execution/AGENTS.md
+++ b/plugins/_code_execution/AGENTS.md
@@ -14,6 +14,7 @@
## Local Contracts
- Keep session concurrency, timeout, streaming, and reset behavior predictable.
+- Terminal reset/close must not hang on foreground commands or shells that ignore SIGTERM.
- Explicitly target local versus SSH execution runtimes.
- Do not hardcode secrets, SSH credentials, or local user paths.
diff --git a/plugins/_code_execution/helpers/tty_session.py b/plugins/_code_execution/helpers/tty_session.py
index 986e63ecd..5d3e3d018 100644
--- a/plugins/_code_execution/helpers/tty_session.py
+++ b/plugins/_code_execution/helpers/tty_session.py
@@ -1,10 +1,12 @@
-import asyncio, os, sys, platform, errno
+import asyncio, os, sys, platform, errno, signal
_IS_WIN = platform.system() == "Windows"
if _IS_WIN:
import winpty # pip install pywinpty # type: ignore
import msvcrt
+_CLOSE_TIMEOUT_SECONDS = 2
+
def _reconfigure_stream_errors(stream) -> None:
reconfigure = getattr(stream, "reconfigure", None)
@@ -77,15 +79,16 @@ class TTYSession:
# Terminate the process if it exists
if self._proc:
+ if getattr(self._proc, "returncode", None) is None:
+ self._signal_process(signal.SIGTERM)
try:
- if getattr(self._proc, "returncode", None) is None:
- self._proc.terminate()
- except ProcessLookupError:
- pass
- except Exception:
- pass
- try:
- await self._proc.wait()
+ await asyncio.wait_for(self._proc.wait(), _CLOSE_TIMEOUT_SECONDS)
+ except asyncio.TimeoutError:
+ self._signal_process(signal.SIGKILL)
+ try:
+ await asyncio.wait_for(self._proc.wait(), _CLOSE_TIMEOUT_SECONDS)
+ except Exception:
+ pass
except Exception:
pass
@@ -93,6 +96,28 @@ class TTYSession:
self._proc = None
self._pump_task = None
+ def _signal_process(self, sig):
+ if self._proc is None:
+ return
+ try:
+ if _IS_WIN:
+ if sig == signal.SIGKILL:
+ self._proc.kill()
+ else:
+ self._proc.terminate()
+ return
+ os.killpg(self._proc.pid, sig)
+ except ProcessLookupError:
+ pass
+ except Exception:
+ try:
+ if sig == signal.SIGKILL:
+ self._proc.kill()
+ else:
+ self._proc.terminate()
+ except Exception:
+ pass
+
def _release_pty_master(self):
"""Release the POSIX PTY master exactly once.
@@ -166,11 +191,7 @@ class TTYSession:
# Only attempt to kill if the process is still running
if getattr(self._proc, "returncode", None) is None:
- try:
- self._proc.kill()
- except ProcessLookupError:
- # Child already gone – treat as successfully killed
- pass
+ self._signal_process(signal.SIGKILL)
self._release_pty_master()
async def read(self, timeout=None):
@@ -241,6 +262,7 @@ async def _spawn_posix_pty(cmd, cwd, env, echo):
cwd=cwd,
env=env,
close_fds=True,
+ start_new_session=True,
)
os.close(slave)
diff --git a/tests/test_code_execution_pager.py b/tests/test_code_execution_pager.py
index 3e80d457d..a4f2d2325 100644
--- a/tests/test_code_execution_pager.py
+++ b/tests/test_code_execution_pager.py
@@ -1,10 +1,13 @@
-"""Regression tests for issue #1697.
+"""Regression tests for code execution shell lifecycle behavior.
Pagers (more/less) must be disabled in the non-interactive shells created by the
code execution tool: without user input they block forever and spin at 100% CPU.
"""
+import asyncio
+
from plugins._code_execution.helpers import shell_local, shell_ssh
+from plugins._code_execution.helpers.tty_session import TTYSession
def test_local_env_disables_pagers_and_preserves_existing():
@@ -30,3 +33,13 @@ def test_local_env_does_not_mutate_input():
def test_ssh_command_disables_pagers():
assert "GIT_PAGER=cat" in shell_ssh.PAGER_DISABLE_COMMAND
assert "PAGER=cat" in shell_ssh.PAGER_DISABLE_COMMAND
+
+
+def test_tty_close_kills_term_resistant_process():
+ async def run():
+ session = TTYSession("bash -lc 'trap \"\" TERM; sleep 30'")
+ await session.start()
+ await asyncio.wait_for(session.close(), timeout=6)
+ assert session._proc is None
+
+ asyncio.run(run())
From f64b6490c5e8eaf10a5087209d9a9164983b4e1d Mon Sep 17 00:00:00 2001
From: Alessandro <155005371+3clyp50@users.noreply.github.com>
Date: Thu, 9 Jul 2026 13:18:06 +0200
Subject: [PATCH 34/43] Add bundled Orchestrator plugin
Promote the terminal agent work into the built-in plugins tree as _orchestrator with Orchestrator branding and the orchestrator skill.
Include adapter status APIs, settings UI, per-agent skill references, thumbnail asset, tests, and DOX coverage for the bundled plugin.
---
plugins/AGENTS.md | 1 +
plugins/_orchestrator/.gitignore | 5 +
plugins/_orchestrator/AGENTS.md | 64 +++
plugins/_orchestrator/LICENSE | 21 +
plugins/_orchestrator/README.md | 187 +++++++
plugins/_orchestrator/api/AGENTS.md | 42 ++
plugins/_orchestrator/api/__init__.py | 0
plugins/_orchestrator/api/disconnect.py | 21 +
.../_orchestrator/api/poll_device_login.py | 16 +
.../_orchestrator/api/start_device_login.py | 18 +
plugins/_orchestrator/api/status.py | 39 ++
plugins/_orchestrator/default_config.yaml | 34 ++
plugins/_orchestrator/helpers/AGENTS.md | 37 ++
plugins/_orchestrator/helpers/__init__.py | 0
.../_orchestrator/helpers/adapters/AGENTS.md | 62 +++
.../helpers/adapters/__init__.py | 0
plugins/_orchestrator/helpers/adapters/a0.py | 65 +++
.../_orchestrator/helpers/adapters/base.py | 63 +++
.../_orchestrator/helpers/adapters/claude.py | 61 +++
.../_orchestrator/helpers/adapters/codex.py | 234 ++++++++
.../_orchestrator/helpers/adapters/cursor.py | 76 +++
.../_orchestrator/helpers/adapters/grok.py | 97 ++++
.../_orchestrator/helpers/adapters/hermes.py | 122 +++++
.../helpers/adapters/opencode.py | 74 +++
plugins/_orchestrator/helpers/registry.py | 44 ++
plugins/_orchestrator/plugin.yaml | 8 +
plugins/_orchestrator/skills/AGENTS.md | 42 ++
.../skills/orchestrator/AGENTS.md | 48 ++
.../skills/orchestrator/SKILL.md | 105 ++++
.../skills/orchestrator/references/AGENTS.md | 34 ++
.../skills/orchestrator/references/a0.md | 31 ++
.../skills/orchestrator/references/claude.md | 50 ++
.../skills/orchestrator/references/codex.md | 36 ++
.../skills/orchestrator/references/cursor.md | 54 ++
.../skills/orchestrator/references/grok.md | 57 ++
.../skills/orchestrator/references/hermes.md | 25 +
.../orchestrator/references/opencode.md | 24 +
plugins/_orchestrator/tests/AGENTS.md | 34 ++
.../tests/test_status_adapters.py | 230 ++++++++
plugins/_orchestrator/thumbnail.png | Bin 0 -> 16585 bytes
plugins/_orchestrator/webui/AGENTS.md | 44 ++
plugins/_orchestrator/webui/config.html | 504 ++++++++++++++++++
.../_orchestrator/webui/orchestrator-store.js | 166 ++++++
43 files changed, 2875 insertions(+)
create mode 100644 plugins/_orchestrator/.gitignore
create mode 100644 plugins/_orchestrator/AGENTS.md
create mode 100644 plugins/_orchestrator/LICENSE
create mode 100644 plugins/_orchestrator/README.md
create mode 100644 plugins/_orchestrator/api/AGENTS.md
create mode 100644 plugins/_orchestrator/api/__init__.py
create mode 100644 plugins/_orchestrator/api/disconnect.py
create mode 100644 plugins/_orchestrator/api/poll_device_login.py
create mode 100644 plugins/_orchestrator/api/start_device_login.py
create mode 100644 plugins/_orchestrator/api/status.py
create mode 100644 plugins/_orchestrator/default_config.yaml
create mode 100644 plugins/_orchestrator/helpers/AGENTS.md
create mode 100644 plugins/_orchestrator/helpers/__init__.py
create mode 100644 plugins/_orchestrator/helpers/adapters/AGENTS.md
create mode 100644 plugins/_orchestrator/helpers/adapters/__init__.py
create mode 100644 plugins/_orchestrator/helpers/adapters/a0.py
create mode 100644 plugins/_orchestrator/helpers/adapters/base.py
create mode 100644 plugins/_orchestrator/helpers/adapters/claude.py
create mode 100644 plugins/_orchestrator/helpers/adapters/codex.py
create mode 100644 plugins/_orchestrator/helpers/adapters/cursor.py
create mode 100644 plugins/_orchestrator/helpers/adapters/grok.py
create mode 100644 plugins/_orchestrator/helpers/adapters/hermes.py
create mode 100644 plugins/_orchestrator/helpers/adapters/opencode.py
create mode 100644 plugins/_orchestrator/helpers/registry.py
create mode 100644 plugins/_orchestrator/plugin.yaml
create mode 100644 plugins/_orchestrator/skills/AGENTS.md
create mode 100644 plugins/_orchestrator/skills/orchestrator/AGENTS.md
create mode 100644 plugins/_orchestrator/skills/orchestrator/SKILL.md
create mode 100644 plugins/_orchestrator/skills/orchestrator/references/AGENTS.md
create mode 100644 plugins/_orchestrator/skills/orchestrator/references/a0.md
create mode 100644 plugins/_orchestrator/skills/orchestrator/references/claude.md
create mode 100644 plugins/_orchestrator/skills/orchestrator/references/codex.md
create mode 100644 plugins/_orchestrator/skills/orchestrator/references/cursor.md
create mode 100644 plugins/_orchestrator/skills/orchestrator/references/grok.md
create mode 100644 plugins/_orchestrator/skills/orchestrator/references/hermes.md
create mode 100644 plugins/_orchestrator/skills/orchestrator/references/opencode.md
create mode 100644 plugins/_orchestrator/tests/AGENTS.md
create mode 100644 plugins/_orchestrator/tests/test_status_adapters.py
create mode 100644 plugins/_orchestrator/thumbnail.png
create mode 100644 plugins/_orchestrator/webui/AGENTS.md
create mode 100644 plugins/_orchestrator/webui/config.html
create mode 100644 plugins/_orchestrator/webui/orchestrator-store.js
diff --git a/plugins/AGENTS.md b/plugins/AGENTS.md
index 768a203b0..c5f604708 100644
--- a/plugins/AGENTS.md
+++ b/plugins/AGENTS.md
@@ -84,6 +84,7 @@ Direct child DOX files:
| [_oauth/AGENTS.md](_oauth/AGENTS.md) | OAuth-backed model-provider connections and local proxy routes. |
| [_office/AGENTS.md](_office/AGENTS.md) | LibreOffice office artifacts and office canvas sessions. |
| [_onboarding/AGENTS.md](_onboarding/AGENTS.md) | First-time model onboarding wizard. |
+| [_orchestrator/AGENTS.md](_orchestrator/AGENTS.md) | External terminal coding-agent orchestration skill, adapter status, and settings UI. |
| [_plugin_installer/AGENTS.md](_plugin_installer/AGENTS.md) | Plugin install and update flows from ZIP, Git, and Plugin Index. |
| [_plugin_scan/AGENTS.md](_plugin_scan/AGENTS.md) | LLM-guided security scanner for third-party plugins. |
| [_plugin_validator/AGENTS.md](_plugin_validator/AGENTS.md) | Plugin manifest, structure, convention, and security validator. |
diff --git a/plugins/_orchestrator/.gitignore b/plugins/_orchestrator/.gitignore
new file mode 100644
index 000000000..c431750d4
--- /dev/null
+++ b/plugins/_orchestrator/.gitignore
@@ -0,0 +1,5 @@
+config.json
+data/
+__pycache__/
+*.py[cod]
+.pytest_cache/
diff --git a/plugins/_orchestrator/AGENTS.md b/plugins/_orchestrator/AGENTS.md
new file mode 100644
index 000000000..7459089f0
--- /dev/null
+++ b/plugins/_orchestrator/AGENTS.md
@@ -0,0 +1,64 @@
+# Orchestrator Plugin - AGENTS.md
+
+## Purpose
+
+- Provide a bundled Agent Zero plugin for delegating repository and coding work to external terminal/headless agents.
+- Keep heavy delegation instructions out of the always-loaded prompt by exposing the `orchestrator` skill instead of a `terminal_agent` tool.
+- Own adapter status metadata, settings UI, Codex device login APIs, and skill instructions for Agent Zero headless, Codex CLI, Claude Code, Cursor CLI, Grok Build, Hermes Agent, OpenCode, and future terminal agents.
+
+## Ownership
+
+- Source-owned files are `.gitignore`, `plugin.yaml`, `default_config.yaml`, `LICENSE`, `README.md`, `thumbnail.png`, `api/`, `helpers/`, `skills/`, `tests/`, and `webui/`.
+- Runtime/user-state files such as `config.json`, `data/`, `__pycache__/`, credential stores, and generated screenshots are not source contracts.
+
+## Local Contracts
+
+- Plugin imports must use `plugins._orchestrator...`.
+- The plugin must remain toolless: no `tools/terminal_agent.py`, no subprocess runner, and no `agent.system.tool.terminal_agent.md` prompt.
+- External agents are orchestrated through the ordinary Agent Zero shell/code execution tool after the `orchestrator` skill is loaded.
+- Adapters report installation, binary resolution, auth state, safe disconnect capability, and optional device-login capability. They do not build or run task commands.
+- Settings UI is for status and command defaults. It must not offer generic install buttons; install/login belongs to the skill-guided human-in-the-loop shell workflow.
+- Credentials must never be hardcoded or printed. Plugin-owned state belongs under `data/<adapter>/` and external CLI state stays in the CLI's own home/config paths.
+- Agent Zero headless is the first adapter. Its setup flow is the exception: ask whether to use this same Agent Zero instance or another/spun-up instance when the target is not specified.
+
+## Work Guidance
+
+- Keep source docs, UI labels, adapter metadata, tests, `skills/orchestrator/SKILL.md`, and per-agent references aligned whenever adding or changing an adapter.
+- Prefer direct, discoverable adapter metadata over a second abstraction layer. Keep generic behavior in the skill and per-agent command details in reference files.
+- For CLI login flows, show provider/menu choices in chat and send the user's selected option back to the active shell session. Do not send users to a Docker shell only to choose a menu.
+- After changing the skill, tell testers to start a fresh chat or reload `orchestrator`; previously loaded skill text can remain attached to old conversations.
+- Codex device login is the only settings-screen OAuth flow currently owned here. Add another only when the credential store and revocation/disconnect path are known.
+- Avoid recursive Agent Zero delegation loops. A0 headless prompts must tell the target instance to answer directly and not delegate back through terminal agents.
+
+## Verification
+
+- Framework-runtime adapter self-check:
+ ```bash
+ docker exec 8dc967046cda bash -lc 'cd /a0 && /opt/venv-a0/bin/python plugins/_orchestrator/tests/test_status_adapters.py'
+ ```
+- Frontend syntax check when editing WebUI JavaScript:
+ ```bash
+ node --check plugins/_orchestrator/webui/orchestrator-store.js
+ ```
+- After Python checks, remove generated bytecode before packaging:
+ ```bash
+ find plugins/_orchestrator -type d -name __pycache__ -prune -exec rm -rf {} +
+ ```
+
+## Child DOX Index
+
+| Child | Scope |
+| --- | --- |
+| [api/AGENTS.md](api/AGENTS.md) | Plugin HTTP API handlers for status, device login polling, and disconnect. |
+| [helpers/AGENTS.md](helpers/AGENTS.md) | Adapter registry, base contracts, and shared helper behavior. |
+| [skills/AGENTS.md](skills/AGENTS.md) | Plugin skill collection and agent-facing delegation instructions. |
+| [tests/AGENTS.md](tests/AGENTS.md) | Direct regression checks for plugin contracts and skill text. |
+| [webui/AGENTS.md](webui/AGENTS.md) | Settings UI markup, Alpine store, status display, and device-login UI. |
+
+Intentionally unindexed local/generated paths:
+
+| Path | Reason |
+| --- | --- |
+| `config.json` | Local plugin settings state. |
+| `data/` | Plugin-owned credential/state directory. |
+| `__pycache__/` | Generated Python bytecode. |
diff --git a/plugins/_orchestrator/LICENSE b/plugins/_orchestrator/LICENSE
new file mode 100644
index 000000000..fdd5c49f3
--- /dev/null
+++ b/plugins/_orchestrator/LICENSE
@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Orchestrator plugin contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
diff --git a/plugins/_orchestrator/README.md b/plugins/_orchestrator/README.md
new file mode 100644
index 000000000..4e1059c08
--- /dev/null
+++ b/plugins/_orchestrator/README.md
@@ -0,0 +1,187 @@
+# Orchestrator
+
+Load-on-demand guidance for orchestrating external **terminal coding agents**
+from Agent Zero without keeping a large delegation tool in every agent prompt.
+
+The plugin provides:
+
+- an `orchestrator` skill that tells Agent Zero how to use external CLIs
+ through the user's A0 CLI host bridge or the normal container shell;
+- a Settings > External Services status screen for configured binaries and
+ detected auth state;
+- adapter metadata for Agent Zero headless, OpenAI Codex CLI, Claude Code,
+ Cursor CLI, Grok Build, Hermes Agent, OpenCode, and future terminal agents.
+
+There is intentionally no `terminal_agent` tool and no settings-screen install
+button. When the user explicitly asks for a terminal agent, the skill first
+decides whether to use the user's own host CLI through A0 CLI or a pal agent
+inside the Agent Zero container.
+
+Host CLI is the primary everyday flow: the user's Claude Code, Codex, Cursor
+CLI, and other coding agents stay installed and logged in on their own
+computer. Container agents remain useful when the user explicitly wants the
+Docker runtime.
+
+## Supported Agents
+
+| Agent | CLI | Login |
+| --- | --- | --- |
+| Agent Zero (headless) | `a0` | Instance `/login` session; `A0_USERNAME`/`A0_PASSWORD` in the shell for protected hosts |
+| OpenAI Codex | `codex` | ChatGPT device login from settings or external CLI login |
+| Claude Code | `claude` | External `claude` login or `ANTHROPIC_API_KEY` |
+| Cursor CLI | `agent` | `CURSOR_API_KEY`, `NO_OPEN_BROWSER=1 agent login`, or cached Cursor login |
+| Grok Build | `grok` | `XAI_API_KEY`, `grok login --device-auth`, or cached Grok login |
+| Hermes Agent | `hermes` | External Hermes/provider setup, `~/.hermes/.env`, `~/.hermes/auth.json`, or provider env vars |
+| OpenCode | `opencode` | External `opencode auth login`, provider env vars, or `~/.local/share/opencode/auth.json` |
+
+The `a0` adapter targets the local Agent Zero instance
+(`http://localhost:80` inside the container) when no `host`/`AGENT_ZERO_HOST`
+is configured, or another Agent Zero instance when you set a host. This lets A0
+delegate to another runtime, and also lets the same runtime ask itself a focused
+question as a secondary use case. A0 is the setup exception: if the user did
+not specify a target, Agent Zero should ask whether to use the same instance or
+another/spun-up instance instead of running a generic login flow. The skill
+tells the target A0 instance to answer directly and not delegate back through
+terminal agents.
+
+## Skill Usage
+
+Ask Agent Zero to load/use the `orchestrator` skill before delegating coding
+work to one of these CLIs. The skill keeps the global setup loop in
+`skills/orchestrator/SKILL.md` and reads one per-agent reference from
+`skills/orchestrator/references/` before acting.
+
+If the user does not specify host versus container, Agent Zero asks and saves a
+per-agent preference with `memory_save`. Local/host mode uses the connector's
+`code_execution_remote` tool and may load `host-code-execution`; container mode
+uses `code_execution_tool`.
+
+For local/host coding agents, the user should run A0 CLI on their own machine:
+
+```bash
+# macOS / Linux
+curl -LsSf https://cli.agent-zero.ai/install.sh | sh
+```
+
+```powershell
+# Windows PowerShell
+irm https://cli.agent-zero.ai/install.ps1 | iex
+```
+
+If A0 CLI is installed but not running, open a terminal and run `a0`, connect it
+to the Agent Zero instance, choose the chat/session or enter the remote/VPS URL,
+then press `F4` for Remote Code Execution. Press `F3` too when host file writes
+are needed. After that, asking Agent Zero to use local Claude Code/Codex/etc.
+will run commands on the host machine instead of inside Docker.
+
+The skill must not use Computer Use to drive coding-agent terminals or TUIs.
+It uses headless CLI commands. The ACP community plugin can be a separate path
+when explicitly requested or when direct CLI automation is unsuitable.
+
+The reference files contain the copy-ready commands, for example Codex:
+
+```bash
+cd "$WORKDIR"
+codex exec --skip-git-repo-check --dangerously-bypass-approvals-and-sandbox "$TASK"
+```
+
+Claude Code defaults to skip permissions for non-interactive runs:
+
+```bash
+cd "$WORKDIR"
+if [ "$(id -u)" -eq 0 ]; then
+ claude -p "$TASK" --output-format json
+else
+ claude -p "$TASK" --output-format json --permission-mode bypassPermissions --allowedTools Bash,Read,Edit
+fi
+```
+
+When Claude Code is not authenticated, use `claude auth login` for the
+human-in-the-loop login step. Do not start plain `claude`; that opens the
+first-run TUI and can trap the agent in theme/provider menus.
+Ask the user to choose the auth mode first: Claude subscription
+(`--claudeai`), Anthropic Console/API billing (`--console`), SSO (`--sso`), or
+an externally set `ANTHROPIC_API_KEY`. In Agent Zero, the user can add that
+key from **Settings > External Services > Secrets Management**; the runtime
+file is `/a0/usr/.env`, not the workdir `.env`.
+If a plain `claude` TUI is already open, reset the terminal session instead of
+sending Enter or `/login`, then use one of the explicit auth commands.
+
+If a non-A0 CLI is missing or asks for login/browser confirmation, Agent Zero
+can install the requested CLI, start its setup/login command, relay the exact
+human step, wait for the user to confirm, then retry a smoke prompt before
+running the real task. If setup shows provider/menu choices, Agent Zero should
+show those options in chat and ask which one to select, not send the user to a
+Docker shell just to choose from a menu. It should not ask the user to paste
+secrets into chat unless there is no safer path.
+
+## Login And Status
+
+Open **Settings > External Services > Orchestrator** to see registered
+adapters, binary paths, installed status, and detected auth source. The screen
+can refresh status and disconnect credentials only when an adapter can safely
+remove a known credential store.
+
+Codex still supports plugin-owned device-code login from the settings screen.
+Those tokens are stored under the plugin data directory (`data/codex/auth.json`,
+mode 600), and the skill can point Codex at that home when needed. If you have
+already logged in with `codex login`, the status screen detects the external
+credentials too.
+
+> Tokens are password-equivalent credentials. Never share one rotating
+> refresh-token auth file between two clients; plugin-owned Codex credentials
+> stay separate from other tools.
+
+## Configuration (`default_config.yaml`)
+
+```yaml
+a0:
+ binary: a0 # falls back to /opt/venv/bin/a0 in Agent Zero Docker
+ host: "" # empty = AGENT_ZERO_HOST env, else local instance
+codex:
+ binary: codex
+ model: ""
+ bypass_sandbox: true
+claude:
+ binary: claude
+ model: ""
+ permission_mode: bypassPermissions
+ allowed_tools: "Bash,Read,Edit"
+ bare: false
+cursor:
+ binary: agent
+ output_format: text
+ force: true
+grok:
+ binary: grok
+ model: ""
+ output_format: json
+ always_approve: true
+ no_auto_update: true
+hermes:
+ binary: hermes
+ model: ""
+ provider: ""
+ toolsets: ""
+ yolo: true
+opencode:
+ binary: opencode
+ model: ""
+ agent: ""
+ auto: true
+```
+
+## Adding A New Agent
+
+1. Create `helpers/adapters/<name>.py` subclassing `TerminalAgentAdapter`
+ (`helpers/adapters/base.py`) and implement `auth_status()`. Add optional
+ safe disconnect or device-login support only when the credential store is
+ known.
+2. Register an instance in `helpers/registry.py`.
+3. Add a config block in `default_config.yaml`.
+4. Add concise command guidance to
+ `skills/orchestrator/references/<name>.md`.
+5. Add the reference to `skills/orchestrator/SKILL.md` and update generic
+ skill guidance only when the global orchestration loop changes.
+
+The status API and settings UI pick registered adapters up automatically.
diff --git a/plugins/_orchestrator/api/AGENTS.md b/plugins/_orchestrator/api/AGENTS.md
new file mode 100644
index 000000000..286103533
--- /dev/null
+++ b/plugins/_orchestrator/api/AGENTS.md
@@ -0,0 +1,42 @@
+# Orchestrator API - AGENTS.md
+
+## Purpose
+
+- Expose small plugin HTTP handlers used by the Orchestrator settings screen.
+- Report registered adapter status and manage only supported, safe auth operations.
+
+## Ownership
+
+- Owns `status.py`, `start_device_login.py`, `poll_device_login.py`, `disconnect.py`, and package initialization for plugin APIs.
+- Does not own execution of terminal agents, installation of CLIs, or generic credential collection.
+
+## Local Contracts
+
+- API handlers derive from `helpers.api.ApiHandler`.
+- Read plugin settings through `helpers.plugins.get_plugin_config("_orchestrator")`.
+- Use `plugins._orchestrator.helpers.registry` for adapter lookup and per-adapter config extraction.
+- Return JSON dictionaries with `ok` and enough context for the WebUI to show a useful error. Do not let adapter exceptions escape to the user as framework failures.
+- `status` may include `install_hint` for display/help only; it must not trigger installation.
+- Device-login endpoints must reject adapters that do not explicitly support device login.
+- `disconnect` may remove only known credential stores or invoke a known safe CLI logout. It must not delete broad home directories or provider config trees.
+
+## Work Guidance
+
+- Keep response shapes stable for `webui/orchestrator-store.js`.
+- When adding adapter auth behavior, update `status.py` only if the shared response needs a new field.
+- Never return token values, API keys, refresh tokens, or raw credential JSON.
+
+## Verification
+
+- Run the plugin self-check after API changes:
+ ```bash
+ docker exec 8dc967046cda bash -lc 'cd /a0 && /opt/venv-a0/bin/python plugins/_orchestrator/tests/test_status_adapters.py'
+ ```
+- For import-only API changes, use the Agent Zero framework runtime:
+ ```bash
+ docker exec 8dc967046cda bash -lc 'cd /a0 && /opt/venv-a0/bin/python -m py_compile plugins/_orchestrator/api/*.py'
+ ```
+
+## Child DOX Index
+
+This folder has no child DOX files.
diff --git a/plugins/_orchestrator/api/__init__.py b/plugins/_orchestrator/api/__init__.py
new file mode 100644
index 000000000..e69de29bb
diff --git a/plugins/_orchestrator/api/disconnect.py b/plugins/_orchestrator/api/disconnect.py
new file mode 100644
index 000000000..c8c0f14be
--- /dev/null
+++ b/plugins/_orchestrator/api/disconnect.py
@@ -0,0 +1,21 @@
+from __future__ import annotations
+
+from helpers.api import ApiHandler, Request
+from helpers.plugins import get_plugin_config
+
+from plugins._orchestrator.helpers.registry import adapter_config, get_adapter
+
+PLUGIN_NAME = "_orchestrator"
+
+
+class Disconnect(ApiHandler):
+ async def process(self, input: dict, request: Request) -> dict:
+ agent_id = str(input.get("agent_id") or "codex")
+ try:
+ adapter = get_adapter(agent_id)
+ plugin_config = get_plugin_config(PLUGIN_NAME) or {}
+ cfg = adapter_config(adapter.id, plugin_config)
+ result = adapter.disconnect(cfg)
+ return {"ok": True, "agent_id": adapter.id, **result}
+ except Exception as exc:
+ return {"ok": False, "agent_id": agent_id, "error": str(exc)}
diff --git a/plugins/_orchestrator/api/poll_device_login.py b/plugins/_orchestrator/api/poll_device_login.py
new file mode 100644
index 000000000..9136fd1e2
--- /dev/null
+++ b/plugins/_orchestrator/api/poll_device_login.py
@@ -0,0 +1,16 @@
+from __future__ import annotations
+
+from helpers.api import ApiHandler, Request
+
+from plugins._orchestrator.helpers.registry import get_adapter
+
+
+class PollDeviceLogin(ApiHandler):
+ async def process(self, input: dict, request: Request) -> dict:
+ agent_id = str(input.get("agent_id") or "codex")
+ try:
+ adapter = get_adapter(agent_id)
+ result = adapter.poll_device_login(input)
+ return {"ok": True, "agent_id": adapter.id, **result}
+ except Exception as exc:
+ return {"ok": False, "agent_id": agent_id, "error": str(exc)}
diff --git a/plugins/_orchestrator/api/start_device_login.py b/plugins/_orchestrator/api/start_device_login.py
new file mode 100644
index 000000000..c0f1767e3
--- /dev/null
+++ b/plugins/_orchestrator/api/start_device_login.py
@@ -0,0 +1,18 @@
+from __future__ import annotations
+
+from helpers.api import ApiHandler, Request
+
+from plugins._orchestrator.helpers.registry import get_adapter
+
+
+class StartDeviceLogin(ApiHandler):
+ async def process(self, input: dict, request: Request) -> dict:
+ agent_id = str(input.get("agent_id") or "codex")
+ try:
+ adapter = get_adapter(agent_id)
+ if not adapter.supports_device_login():
+ raise RuntimeError(f"{adapter.title} does not support device login.")
+ result = adapter.start_device_login()
+ return {"ok": True, "agent_id": adapter.id, **result}
+ except Exception as exc:
+ return {"ok": False, "agent_id": agent_id, "error": str(exc)}
diff --git a/plugins/_orchestrator/api/status.py b/plugins/_orchestrator/api/status.py
new file mode 100644
index 000000000..5603e6d15
--- /dev/null
+++ b/plugins/_orchestrator/api/status.py
@@ -0,0 +1,39 @@
+from __future__ import annotations
+
+from helpers.api import ApiHandler, Request
+from helpers.plugins import get_plugin_config
+
+from plugins._orchestrator.helpers.registry import adapter_config, list_adapters
+
+PLUGIN_NAME = "_orchestrator"
+
+
+class Status(ApiHandler):
+ async def process(self, input: dict, request: Request) -> dict:
+ plugin_config = get_plugin_config(PLUGIN_NAME) or {}
+ agents = []
+ for adapter in list_adapters():
+ cfg = adapter_config(adapter.id, plugin_config)
+ try:
+ auth = adapter.auth_status(cfg)
+ except Exception as exc:
+ auth = {
+ "connected": False,
+ "mode": "",
+ "auth_path": "",
+ "error": str(exc),
+ }
+ agents.append(
+ {
+ "id": adapter.id,
+ "title": adapter.title,
+ "description": adapter.description,
+ "binary": adapter.resolve_binary(cfg),
+ "installed": adapter.is_installed(cfg),
+ "install_hint": adapter.install_hint,
+ "supports_device_login": adapter.supports_device_login(),
+ "can_disconnect": adapter.can_disconnect(cfg),
+ "auth": auth,
+ }
+ )
+ return {"ok": True, "agents": agents}
diff --git a/plugins/_orchestrator/default_config.yaml b/plugins/_orchestrator/default_config.yaml
new file mode 100644
index 000000000..d90d6ecc4
--- /dev/null
+++ b/plugins/_orchestrator/default_config.yaml
@@ -0,0 +1,34 @@
+a0:
+ binary: a0 # falls back to /opt/venv/bin/a0 in Agent Zero Docker
+ host: "" # empty = AGENT_ZERO_HOST env, else local instance (http://localhost:80)
+codex:
+ binary: codex
+ model: ""
+ bypass_sandbox: true
+claude:
+ binary: claude
+ model: ""
+ permission_mode: bypassPermissions
+ allowed_tools: "Bash,Read,Edit"
+ bare: false
+cursor:
+ binary: agent
+ output_format: text
+ force: true
+grok:
+ binary: grok
+ model: ""
+ output_format: json
+ always_approve: true
+ no_auto_update: true
+hermes:
+ binary: hermes
+ model: ""
+ provider: ""
+ toolsets: ""
+ yolo: true
+opencode:
+ binary: opencode
+ model: ""
+ agent: ""
+ auto: true
diff --git a/plugins/_orchestrator/helpers/AGENTS.md b/plugins/_orchestrator/helpers/AGENTS.md
new file mode 100644
index 000000000..c5cc9d8e1
--- /dev/null
+++ b/plugins/_orchestrator/helpers/AGENTS.md
@@ -0,0 +1,37 @@
+# Orchestrator Helpers - AGENTS.md
+
+## Purpose
+
+- Provide the status adapter framework and registry for terminal/headless coding agents.
+- Keep adapter logic focused on discovery, configuration, authentication state, and safe credential cleanup.
+
+## Ownership
+
+- Owns `registry.py`, package initialization, and the `adapters/` subtree.
+- Does not own shell command execution, long-running process management, or task output parsing.
+
+## Local Contracts
+
+- `registry.py` is the only place that orders and exposes registered adapters.
+- A0 Headless must remain first in `list_adapters()` unless product direction changes.
+- Registry imports must use `plugins._orchestrator...` for bundled plugin imports.
+- `adapter_config()` must tolerate missing or malformed plugin config and return an empty dict.
+- Add adapters by creating a `TerminalAgentAdapter` subclass, registering an instance in `registry.py`, adding defaults in `default_config.yaml`, and updating the skill, README, and tests.
+
+## Work Guidance
+
+- Keep helpers boring and inspectable. If command orchestration is needed, put instructions in the skill rather than adding a runner.
+- Adapter metadata should be readable enough for both the UI and agent skill documentation to stay in sync.
+
+## Verification
+
+- Run the adapter contract checks:
+ ```bash
+ docker exec 8dc967046cda bash -lc 'cd /a0 && /opt/venv-a0/bin/python plugins/_orchestrator/tests/test_status_adapters.py'
+ ```
+
+## Child DOX Index
+
+| Child | Scope |
+| --- | --- |
+| [adapters/AGENTS.md](adapters/AGENTS.md) | Per-agent adapter contracts, auth detection, and safe disconnect rules. |
diff --git a/plugins/_orchestrator/helpers/__init__.py b/plugins/_orchestrator/helpers/__init__.py
new file mode 100644
index 000000000..e69de29bb
diff --git a/plugins/_orchestrator/helpers/adapters/AGENTS.md b/plugins/_orchestrator/helpers/adapters/AGENTS.md
new file mode 100644
index 000000000..4f852e401
--- /dev/null
+++ b/plugins/_orchestrator/helpers/adapters/AGENTS.md
@@ -0,0 +1,62 @@
+# Terminal Agent Adapters - AGENTS.md
+
+## Purpose
+
+- Represent each supported external terminal agent as status/auth metadata for the plugin UI and APIs.
+- Keep every adapter compatible with the shared `TerminalAgentAdapter` contract.
+
+## Ownership
+
+- Owns `base.py` plus adapter modules for A0 Headless, Codex CLI, Claude Code, Cursor CLI, Grok Build, Hermes Agent, OpenCode, and future status adapters.
+- Owns credential-path detection and safe disconnect behavior only when the adapter can identify the exact credential store.
+
+## Local Contracts
+
+- Subclasses must define `id`, `title`, `binary`, `install_hint`, `description`, and `auth_status()`.
+- `auth_status()` returns a dict containing at least `connected`, `mode`, and `auth_path`; optional `error` must be display-safe.
+- `install_hint` is informational. It must not become executable API behavior.
+- `resolve_binary()` should respect configured absolute paths and fall back to the adapter's default binary.
+- `is_installed()` checks executability only. It must not install, mutate, or prompt.
+- Do not add command execution hooks such as `build_command`, `install_command`, `parse_session_id`, or `format_output`.
+- `data_dir()` is only for plugin-owned private state such as Codex device-login auth. Do not store user-entered provider keys in source or broad config files.
+
+## Work Guidance
+
+- A0 Headless:
+ - Host resolution is config `a0.host`, then `AGENT_ZERO_HOST`, then `http://localhost:80` inside the Agent Zero container.
+ - The Docker fallback binary is `/opt/venv/bin/a0` when plain `a0` is unavailable.
+ - Status means the host socket is reachable; login/target choice is handled by the skill.
+- Codex CLI:
+ - Detect plugin-owned `data/codex/auth.json` before external `CODEX_HOME` or `~/.codex/auth.json`.
+ - Keep device-code OAuth compatible with the Agent Zero `_oauth` reference.
+ - External disconnect may call `codex logout`; plugin-owned disconnect deletes only the plugin auth file.
+- Claude Code:
+ - Treat `ANTHROPIC_API_KEY` as environment auth.
+ - Detect the CLI credentials file but never read or return credential contents.
+ - Do not model the first-run TUI as a status API flow.
+- Cursor CLI:
+ - Treat `CURSOR_API_KEY` as environment auth.
+ - Detect known files under `~/.cursor/` without returning secret contents.
+ - Do not model the interactive terminal UI as a status API flow.
+- Grok Build:
+ - Treat `XAI_API_KEY` as environment auth.
+ - Detect `~/.grok/config.toml`, `~/.grok/auth.json`, and `~/.grok/auth/` without returning secret contents.
+ - Do not model the full-screen TUI as a status API flow.
+- Hermes Agent and OpenCode:
+ - Detect known provider environment variables and known auth files.
+ - Secret detection should answer yes/no without returning secret values.
+
+## Verification
+
+- Run adapter tests:
+ ```bash
+ docker exec 8dc967046cda bash -lc 'cd /a0 && /opt/venv-a0/bin/python plugins/_orchestrator/tests/test_status_adapters.py'
+ ```
+- For syntax-only adapter edits:
+ ```bash
+ docker exec 8dc967046cda bash -lc 'cd /a0 && /opt/venv-a0/bin/python -m py_compile plugins/_orchestrator/helpers/adapters/*.py'
+ ```
+
+## Child DOX Index
+
+This folder has no child DOX files.
diff --git a/plugins/_orchestrator/helpers/adapters/__init__.py b/plugins/_orchestrator/helpers/adapters/__init__.py
new file mode 100644
index 000000000..e69de29bb
diff --git a/plugins/_orchestrator/helpers/adapters/a0.py b/plugins/_orchestrator/helpers/adapters/a0.py
new file mode 100644
index 000000000..b1a46f8ae
--- /dev/null
+++ b/plugins/_orchestrator/helpers/adapters/a0.py
@@ -0,0 +1,65 @@
+from __future__ import annotations
+
+import os
+import shutil
+import socket
+from pathlib import Path
+from typing import Any
+from urllib.parse import urlparse
+
+from plugins._orchestrator.helpers.adapters.base import TerminalAgentAdapter
+
+# Inside the Agent Zero container, the local WebUI listens on port 80,
+# so the default target is the instance running this very plugin.
+DEFAULT_HOST = "http://localhost:80"
+DEFAULT_DOCKER_A0_BINARY = "/opt/venv/bin/a0"
+
+
+class AgentZeroAdapter(TerminalAgentAdapter):
+ """Delegate tasks to an Agent Zero instance via `a0 headless` one-shot mode.
+
+ Host resolution: adapter config `host` > AGENT_ZERO_HOST env > local
+ instance (http://localhost:80 inside the container).
+ """
+
+ id = "a0"
+ title = "Agent Zero (headless)"
+ binary = "a0"
+ install_hint = "pip install git+https://github.com/agent0ai/a0-connector.git@development"
+ description = "Delegate to this or another Agent Zero instance through a0 headless."
+
+ # --- connection ----------------------------------------------------------
+
+ def resolve_binary(self, config: dict[str, Any] | None = None) -> str:
+ binary = super().resolve_binary(config)
+ if binary == self.binary and shutil.which(binary) is None:
+ bundled = Path(DEFAULT_DOCKER_A0_BINARY)
+ if bundled.is_file():
+ return str(bundled)
+ return binary
+
+ def resolve_host(self, config: dict[str, Any] | None = None) -> str:
+ cfg = config or {}
+ host = str(cfg.get("host") or "").strip()
+ if host:
+ return host
+ env_host = os.environ.get("AGENT_ZERO_HOST", "").strip()
+ return env_host or DEFAULT_HOST
+
+ def auth_status(self, config: dict[str, Any] | None = None) -> dict[str, Any]:
+ host = self.resolve_host(config)
+ if _probe(host):
+ return {"connected": True, "mode": "external", "auth_path": host}
+ return {"connected": False, "mode": "", "auth_path": host}
+
+
+def _probe(host_url: str) -> bool:
+ url = host_url if "://" in host_url else f"http://{host_url}"
+ parsed = urlparse(url)
+ host = parsed.hostname or "localhost"
+ port = parsed.port or (443 if parsed.scheme == "https" else 80)
+ try:
+ with socket.create_connection((host, port), timeout=2):
+ return True
+ except OSError:
+ return False
diff --git a/plugins/_orchestrator/helpers/adapters/base.py b/plugins/_orchestrator/helpers/adapters/base.py
new file mode 100644
index 000000000..e6e94e9b7
--- /dev/null
+++ b/plugins/_orchestrator/helpers/adapters/base.py
@@ -0,0 +1,63 @@
+from __future__ import annotations
+
+import os
+import shutil
+from abc import ABC, abstractmethod
+from pathlib import Path
+from typing import Any
+
+from helpers import files
+
+
+class TerminalAgentAdapter(ABC):
+ """Contract every terminal coding agent status adapter must implement."""
+
+ id: str = ""
+ title: str = ""
+ binary: str = ""
+ install_hint: str = ""
+ description: str = ""
+
+ # --- environment -----------------------------------------------------
+
+ def data_dir(self) -> Path:
+ """Plugin-owned private directory for this adapter (auth, state)."""
+ path = Path(
+ files.get_abs_path("usr", "plugins", "_orchestrator", "data", self.id)
+ )
+ path.mkdir(parents=True, exist_ok=True)
+ return path
+
+ def resolve_binary(self, config: dict[str, Any] | None = None) -> str:
+ cfg = config if isinstance(config, dict) else {}
+ configured = str(cfg.get("binary") or "").strip()
+ return configured or self.binary
+
+ def is_installed(self, config: dict[str, Any] | None = None) -> bool:
+ binary = self.resolve_binary(config)
+ if not binary:
+ return False
+ if os.path.isabs(binary):
+ return Path(binary).is_file() and os.access(binary, os.X_OK)
+ return shutil.which(binary) is not None
+
+ # --- authentication ----------------------------------------------------
+
+ @abstractmethod
+ def auth_status(self, config: dict[str, Any] | None = None) -> dict[str, Any]:
+ """Return {connected: bool, mode: 'plugin'|'external'|'', auth_path: str}."""
+
+ def supports_device_login(self) -> bool:
+ return False
+
+ def start_device_login(self) -> dict[str, Any]:
+ raise NotImplementedError(f"{self.id} does not support device login.")
+
+ def poll_device_login(self, payload: dict[str, Any]) -> dict[str, Any]:
+ raise NotImplementedError(f"{self.id} does not support device login.")
+
+ def can_disconnect(self, config: dict[str, Any] | None = None) -> bool:
+ return False
+
+ def disconnect(self, config: dict[str, Any] | None = None) -> dict[str, Any]:
+ raise NotImplementedError(f"{self.id} does not support disconnect.")
diff --git a/plugins/_orchestrator/helpers/adapters/claude.py b/plugins/_orchestrator/helpers/adapters/claude.py
new file mode 100644
index 000000000..987faebfb
--- /dev/null
+++ b/plugins/_orchestrator/helpers/adapters/claude.py
@@ -0,0 +1,61 @@
+from __future__ import annotations
+
+import os
+from pathlib import Path
+from typing import Any
+
+from plugins._orchestrator.helpers.adapters.base import TerminalAgentAdapter
+
+
+class ClaudeCodeAdapter(TerminalAgentAdapter):
+ id = "claude"
+ title = "Claude Code"
+ binary = "claude"
+ install_hint = "npm install -g @anthropic-ai/claude-code"
+ description = "Anthropic Claude Code CLI in non-interactive print mode."
+
+ def _credentials_path(self) -> Path:
+ config_dir = os.environ.get("CLAUDE_CONFIG_DIR", "").strip()
+ root = Path(config_dir).expanduser() if config_dir else Path.home() / ".claude"
+ return root / ".credentials.json"
+
+ def auth_status(self, config: dict[str, Any] | None = None) -> dict[str, Any]:
+ if os.environ.get("ANTHROPIC_API_KEY"):
+ return {
+ "connected": True,
+ "mode": "env",
+ "auth_path": "ANTHROPIC_API_KEY",
+ }
+ credentials = self._credentials_path()
+ try:
+ if credentials.is_file() and credentials.stat().st_size > 0:
+ return {
+ "connected": True,
+ "mode": "external",
+ "auth_path": str(credentials),
+ }
+ except OSError as exc:
+ return {
+ "connected": False,
+ "mode": "",
+ "auth_path": "",
+ "error": str(exc),
+ }
+ if not self.is_installed(config):
+ return {"connected": False, "mode": "", "auth_path": ""}
+ return {"connected": False, "mode": "", "auth_path": ""}
+
+ def can_disconnect(self, config: dict[str, Any] | None = None) -> bool:
+ return not os.environ.get("ANTHROPIC_API_KEY") and self._credentials_path().is_file()
+
+ def disconnect(self, config: dict[str, Any] | None = None) -> dict[str, Any]:
+ if os.environ.get("ANTHROPIC_API_KEY"):
+ return {
+ "removed": False,
+ "message": "ANTHROPIC_API_KEY is set in the environment.",
+ }
+ path = self._credentials_path()
+ if path.exists():
+ path.unlink()
+ return {"removed": True, "mode": "external"}
+ return {"removed": False, "message": "No Claude Code credentials found."}
diff --git a/plugins/_orchestrator/helpers/adapters/codex.py b/plugins/_orchestrator/helpers/adapters/codex.py
new file mode 100644
index 000000000..c65adf516
--- /dev/null
+++ b/plugins/_orchestrator/helpers/adapters/codex.py
@@ -0,0 +1,234 @@
+from __future__ import annotations
+
+import json
+import os
+import subprocess
+import time
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any
+
+import requests
+
+from plugins._orchestrator.helpers.adapters.base import TerminalAgentAdapter
+
+# Codex CLI's public OAuth client (same constants the official CLI uses;
+# see also plugins/_oauth for the reference implementation).
+CLIENT_ID = "app_EMoamEEZ73f0CkXaXp7hrann"
+ISSUER = "https://auth.openai.com"
+TOKEN_URL = "https://auth.openai.com/oauth/token"
+AUTH_FILENAME = "auth.json"
+DEVICE_CODE_TIMEOUT_SECONDS = 15 * 60
+
+
+class CodexAdapter(TerminalAgentAdapter):
+ id = "codex"
+ title = "OpenAI Codex CLI"
+ binary = "codex"
+ install_hint = "npm install -g @openai/codex"
+ description = "OpenAI Codex CLI for autonomous coding tasks in a workdir."
+
+ # --- authentication ----------------------------------------------------
+
+ def _plugin_auth_path(self) -> Path:
+ return self.data_dir() / AUTH_FILENAME
+
+ def _external_auth_path(self) -> Path:
+ codex_home = os.environ.get("CODEX_HOME", "").strip()
+ if codex_home:
+ return Path(codex_home).expanduser() / AUTH_FILENAME
+ return Path.home() / ".codex" / AUTH_FILENAME
+
+ def auth_status(self, config: dict[str, Any] | None = None) -> dict[str, Any]:
+ plugin_path = self._plugin_auth_path()
+ if _has_chatgpt_tokens(plugin_path):
+ return {
+ "connected": True,
+ "mode": "plugin",
+ "auth_path": str(plugin_path),
+ }
+ external_path = self._external_auth_path()
+ if _has_chatgpt_tokens(external_path):
+ return {
+ "connected": True,
+ "mode": "external",
+ "auth_path": str(external_path),
+ }
+ return {"connected": False, "mode": "", "auth_path": ""}
+
+ def supports_device_login(self) -> bool:
+ return True
+
+ def start_device_login(self) -> dict[str, Any]:
+ response = requests.post(
+ f"{ISSUER}/api/accounts/deviceauth/usercode",
+ headers={"Content-Type": "application/json"},
+ json={"client_id": CLIENT_ID},
+ timeout=30,
+ )
+ if not response.ok:
+ raise RuntimeError(_oauth_error(response))
+ payload = response.json() if isinstance(response.json(), dict) else {}
+ device_auth_id = str(payload.get("device_auth_id") or "")
+ user_code = str(payload.get("user_code") or payload.get("usercode") or "")
+ if not device_auth_id or not user_code:
+ raise RuntimeError("Device authorization response did not include a code.")
+ return {
+ "device_auth_id": device_auth_id,
+ "user_code": user_code,
+ "interval": _safe_int(payload.get("interval"), 5),
+ "expires_at": time.time() + DEVICE_CODE_TIMEOUT_SECONDS,
+ "verification_url": f"{ISSUER}/codex/device",
+ }
+
+ def poll_device_login(self, payload: dict[str, Any]) -> dict[str, Any]:
+ device_auth_id = str(payload.get("device_auth_id") or "")
+ user_code = str(payload.get("user_code") or "")
+ if not device_auth_id or not user_code:
+ raise RuntimeError("Missing device_auth_id or user_code.")
+ response = requests.post(
+ f"{ISSUER}/api/accounts/deviceauth/token",
+ headers={"Content-Type": "application/json"},
+ json={"device_auth_id": device_auth_id, "user_code": user_code},
+ timeout=30,
+ )
+ if response.status_code in {403, 404}:
+ return {"completed": False}
+ if not response.ok:
+ raise RuntimeError(_oauth_error(response))
+ data = response.json() if isinstance(response.json(), dict) else {}
+ authorization_code = str(data.get("authorization_code") or "")
+ verifier = str(data.get("code_verifier") or "")
+ if not authorization_code or not verifier:
+ raise RuntimeError("Device authorization response missing exchange data.")
+ tokens = _exchange_code(
+ authorization_code, f"{ISSUER}/deviceauth/callback", verifier
+ )
+ account_id = _derive_account_id(tokens["id_token"])
+ self._write_auth(tokens, account_id)
+ return {"completed": True, "account_id": account_id}
+
+ def can_disconnect(self, config: dict[str, Any] | None = None) -> bool:
+ status = self.auth_status(config)
+ return bool(status.get("connected"))
+
+ def disconnect(self, config: dict[str, Any] | None = None) -> dict[str, Any]:
+ status = self.auth_status(config)
+ path = self._plugin_auth_path()
+ if status.get("mode") == "plugin" and path.exists():
+ path.unlink()
+ return {"removed": True, "mode": "plugin"}
+ if status.get("mode") == "external":
+ result = subprocess.run(
+ [self.resolve_binary(config), "logout"],
+ capture_output=True,
+ text=True,
+ timeout=30,
+ )
+ if result.returncode != 0:
+ raise RuntimeError((result.stderr or result.stdout).strip())
+ return {"removed": True, "mode": "external"}
+ return {"removed": False, "message": "No Codex credentials found."}
+
+ def _write_auth(self, tokens: dict[str, str], account_id: str) -> None:
+ auth_data = {
+ "auth_mode": "chatgpt",
+ "OPENAI_API_KEY": None,
+ "tokens": {
+ "id_token": tokens["id_token"],
+ "access_token": tokens["access_token"],
+ "refresh_token": tokens["refresh_token"],
+ "account_id": account_id,
+ },
+ "last_refresh": _utc_now_iso(),
+ }
+ path = self._plugin_auth_path()
+ path.parent.mkdir(parents=True, exist_ok=True)
+ tmp = path.with_suffix(".tmp")
+ tmp.write_text(json.dumps(auth_data, indent=2))
+ os.chmod(tmp, 0o600)
+ tmp.replace(path)
+
+
+def _has_chatgpt_tokens(path: Path) -> bool:
+ try:
+ data = json.loads(path.read_text())
+ except (OSError, ValueError):
+ return False
+ tokens = data.get("tokens") if isinstance(data, dict) else None
+ if not isinstance(tokens, dict):
+ return False
+ return bool(tokens.get("access_token") and tokens.get("refresh_token"))
+
+
+def _exchange_code(code: str, redirect_uri: str, verifier: str) -> dict[str, str]:
+ response = requests.post(
+ TOKEN_URL,
+ headers={"Content-Type": "application/x-www-form-urlencoded"},
+ data={
+ "grant_type": "authorization_code",
+ "code": code,
+ "redirect_uri": redirect_uri,
+ "client_id": CLIENT_ID,
+ "code_verifier": verifier,
+ },
+ timeout=30,
+ )
+ if not response.ok:
+ raise RuntimeError(_oauth_error(response))
+ payload = response.json() if isinstance(response.json(), dict) else {}
+ tokens = {
+ "id_token": str(payload.get("id_token") or ""),
+ "access_token": str(payload.get("access_token") or ""),
+ "refresh_token": str(payload.get("refresh_token") or ""),
+ }
+ missing = [key for key, value in tokens.items() if not value]
+ if missing:
+ raise RuntimeError(f"OAuth token response is missing: {', '.join(missing)}")
+ return tokens
+
+
+def _derive_account_id(id_token: str) -> str:
+ claims = _parse_jwt_claims(id_token)
+ auth_claims = claims.get("https://api.openai.com/auth")
+ if isinstance(auth_claims, dict):
+ value = auth_claims.get("chatgpt_account_id")
+ if isinstance(value, str):
+ return value
+ return ""
+
+
+def _parse_jwt_claims(token: str) -> dict[str, Any]:
+ import base64
+
+ try:
+ payload_part = token.split(".")[1]
+ padded = payload_part + "=" * (-len(payload_part) % 4)
+ value = json.loads(base64.urlsafe_b64decode(padded))
+ return value if isinstance(value, dict) else {}
+ except Exception:
+ return {}
+
+
+def _oauth_error(response: requests.Response) -> str:
+ try:
+ payload = response.json()
+ except Exception:
+ payload = None
+ if isinstance(payload, dict):
+ for key in ("error_description", "error"):
+ value = payload.get(key)
+ if isinstance(value, str) and value:
+ return value
+ return f"OAuth endpoint returned status {response.status_code}."
+
+
+def _safe_int(value: Any, default: int) -> int:
+ try:
+ return int(value)
+ except (TypeError, ValueError):
+ return default
+
+
+def _utc_now_iso() -> str:
+ return datetime.now(timezone.utc).isoformat().replace("+00:00", "Z")
diff --git a/plugins/_orchestrator/helpers/adapters/cursor.py b/plugins/_orchestrator/helpers/adapters/cursor.py
new file mode 100644
index 000000000..cb7e502bf
--- /dev/null
+++ b/plugins/_orchestrator/helpers/adapters/cursor.py
@@ -0,0 +1,76 @@
+from __future__ import annotations
+
+import json
+import os
+from pathlib import Path
+from typing import Any
+
+from plugins._orchestrator.helpers.adapters.base import TerminalAgentAdapter
+
+
+_SECRET_KEYS = {"access_token", "api_key", "id_token", "refresh_token", "token"}
+_AUTH_FILES = (
+ "auth.json",
+ "credentials.json",
+ "token.json",
+ "agent/auth.json",
+ "agent/credentials.json",
+ "agent/token.json",
+)
+
+
+class CursorCliAdapter(TerminalAgentAdapter):
+ id = "cursor"
+ title = "Cursor CLI"
+ binary = "agent"
+ install_hint = "curl https://cursor.com/install -fsS | bash"
+ description = "Cursor Agent CLI in headless print mode."
+
+ def _home(self) -> Path:
+ configured = os.environ.get("CURSOR_HOME", "").strip()
+ return Path(configured).expanduser() if configured else Path.home() / ".cursor"
+
+ def auth_status(self, config: dict[str, Any] | None = None) -> dict[str, Any]:
+ env_var = next(
+ (name for name in ("CURSOR_API_KEY", "API_KEY_CURSOR") if os.environ.get(name)),
+ "",
+ )
+ if env_var:
+ return {"connected": True, "mode": "env", "auth_path": env_var}
+
+ home = self._home()
+ for relative in _AUTH_FILES:
+ path = home / relative
+ try:
+ if _file_has_secret(path):
+ return {"connected": True, "mode": "external", "auth_path": str(path)}
+ except OSError as exc:
+ return {"connected": False, "mode": "", "auth_path": "", "error": str(exc)}
+
+ return {"connected": False, "mode": "", "auth_path": str(home)}
+
+
+def _file_has_secret(path: Path) -> bool:
+ if not path.is_file() or path.stat().st_size <= 0:
+ return False
+ text = path.read_text(encoding="utf-8", errors="ignore")
+ try:
+ return _contains_secret(json.loads(text))
+ except ValueError:
+ return _text_has_secret(text)
+
+
+def _text_has_secret(text: str) -> bool:
+ return any(f'"{key}"' in text or f"{key} =" in text for key in _SECRET_KEYS)
+
+
+def _contains_secret(value: Any) -> bool:
+ if isinstance(value, dict):
+ for key, item in value.items():
+ if str(key) in _SECRET_KEYS and isinstance(item, str) and item.strip():
+ return True
+ if _contains_secret(item):
+ return True
+ if isinstance(value, list):
+ return any(_contains_secret(item) for item in value)
+ return False
diff --git a/plugins/_orchestrator/helpers/adapters/grok.py b/plugins/_orchestrator/helpers/adapters/grok.py
new file mode 100644
index 000000000..2e3ead1bc
--- /dev/null
+++ b/plugins/_orchestrator/helpers/adapters/grok.py
@@ -0,0 +1,97 @@
+from __future__ import annotations
+
+import json
+import os
+from pathlib import Path
+from typing import Any
+
+from plugins._orchestrator.helpers.adapters.base import TerminalAgentAdapter
+
+
+_SECRET_KEYS = {"access_token", "api_key", "id_token", "refresh_token", "token"}
+
+
+class GrokBuildAdapter(TerminalAgentAdapter):
+ id = "grok"
+ title = "Grok Build"
+ binary = "grok"
+ install_hint = "curl -fsSL https://x.ai/cli/install.sh | bash"
+ description = "xAI Grok Build CLI in headless single-prompt mode."
+
+ def _home(self) -> Path:
+ configured = os.environ.get("GROK_HOME", "").strip()
+ return Path(configured).expanduser() if configured else Path.home() / ".grok"
+
+ def auth_status(self, config: dict[str, Any] | None = None) -> dict[str, Any]:
+ env_var = next(
+ (name for name in ("XAI_API_KEY", "API_KEY_XAI") if os.environ.get(name)),
+ "",
+ )
+ if env_var:
+ return {"connected": True, "mode": "env", "auth_path": env_var}
+
+ home = self._home()
+ for path in (home / "config.toml", home / "auth.json"):
+ try:
+ if _file_has_secret(path):
+ return {"connected": True, "mode": "external", "auth_path": str(path)}
+ except OSError as exc:
+ return {"connected": False, "mode": "", "auth_path": "", "error": str(exc)}
+
+ auth_dir = home / "auth"
+ try:
+ if auth_dir.is_dir():
+ for path in auth_dir.iterdir():
+ if path.is_file() and _file_has_secret(path):
+ return {
+ "connected": True,
+ "mode": "external",
+ "auth_path": str(auth_dir),
+ }
+ except OSError as exc:
+ return {"connected": False, "mode": "", "auth_path": "", "error": str(exc)}
+
+ return {"connected": False, "mode": "", "auth_path": str(home)}
+
+
+def _file_has_secret(path: Path) -> bool:
+ if not path.is_file() or path.stat().st_size <= 0:
+ return False
+ text = path.read_text(encoding="utf-8", errors="ignore")
+ if path.suffix == ".toml":
+ return _toml_has_secret(text)
+ try:
+ return _contains_secret(json.loads(text))
+ except ValueError:
+ return _text_has_secret(text)
+
+
+def _toml_has_secret(text: str) -> bool:
+ for line in text.splitlines():
+ raw = line.strip()
+ if not raw or raw.startswith("#") or "=" not in raw:
+ continue
+ key, value = raw.split("=", 1)
+ key = key.strip()
+ value = value.strip().strip("'\"")
+ if key == "api_key" and value:
+ return True
+ if key == "env_key" and value and os.environ.get(value):
+ return True
+ return False
+
+
+def _text_has_secret(text: str) -> bool:
+ return any(f'"{key}"' in text or f"{key} =" in text for key in _SECRET_KEYS)
+
+
+def _contains_secret(value: Any) -> bool:
+ if isinstance(value, dict):
+ for key, item in value.items():
+ if str(key) in _SECRET_KEYS and isinstance(item, str) and item.strip():
+ return True
+ if _contains_secret(item):
+ return True
+ if isinstance(value, list):
+ return any(_contains_secret(item) for item in value)
+ return False
diff --git a/plugins/_orchestrator/helpers/adapters/hermes.py b/plugins/_orchestrator/helpers/adapters/hermes.py
new file mode 100644
index 000000000..7946fee63
--- /dev/null
+++ b/plugins/_orchestrator/helpers/adapters/hermes.py
@@ -0,0 +1,122 @@
+from __future__ import annotations
+
+import json
+import os
+from pathlib import Path
+from typing import Any
+
+from plugins._orchestrator.helpers.adapters.base import TerminalAgentAdapter
+
+
+_AUTH_ENV_VARS = {
+ "ALIBABA_CODING_PLAN_API_KEY",
+ "ANTHROPIC_API_KEY",
+ "ANTHROPIC_TOKEN",
+ "ARCEEAI_API_KEY",
+ "AZURE_FOUNDRY_API_KEY",
+ "CLAUDE_CODE_OAUTH_TOKEN",
+ "COPILOT_GITHUB_TOKEN",
+ "DASHSCOPE_API_KEY",
+ "DEEPSEEK_API_KEY",
+ "GEMINI_API_KEY",
+ "GH_TOKEN",
+ "GITHUB_TOKEN",
+ "GLM_API_KEY",
+ "GMI_API_KEY",
+ "GOOGLE_API_KEY",
+ "HF_TOKEN",
+ "KILOCODE_API_KEY",
+ "KIMI_API_KEY",
+ "KIMI_CODING_API_KEY",
+ "MINIMAX_API_KEY",
+ "NVIDIA_API_KEY",
+ "NOVITA_API_KEY",
+ "OLLAMA_API_KEY",
+ "OPENAI_API_KEY",
+ "OPENCODE_GO_API_KEY",
+ "OPENCODE_ZEN_API_KEY",
+ "OPENROUTER_API_KEY",
+ "STEPFUN_API_KEY",
+ "TOKENHUB_API_KEY",
+ "XAI_API_KEY",
+ "ZAI_API_KEY",
+}
+
+_SECRET_KEYS = {
+ "access_token",
+ "agent_key",
+ "api_key",
+ "id_token",
+ "refresh_token",
+ "runtime_api_key",
+ "token",
+}
+
+
+class HermesAgentAdapter(TerminalAgentAdapter):
+ id = "hermes"
+ title = "Hermes Agent"
+ binary = "hermes"
+ install_hint = "curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash"
+ description = "Nous Research Hermes Agent CLI in quiet headless chat mode."
+
+ def _home(self) -> Path:
+ configured = os.environ.get("HERMES_HOME", "").strip()
+ return Path(configured).expanduser() if configured else Path.home() / ".hermes"
+
+ def auth_status(self, config: dict[str, Any] | None = None) -> dict[str, Any]:
+ env_var = next((name for name in sorted(_AUTH_ENV_VARS) if os.environ.get(name)), "")
+ if env_var:
+ return {"connected": True, "mode": "env", "auth_path": env_var}
+
+ home = self._home()
+ env_path = home / ".env"
+ try:
+ if _env_file_has_key(env_path):
+ return {"connected": True, "mode": "external", "auth_path": str(env_path)}
+ except OSError as exc:
+ return {"connected": False, "mode": "", "auth_path": "", "error": str(exc)}
+
+ auth_path = home / "auth.json"
+ try:
+ if _auth_store_has_credentials(auth_path):
+ return {"connected": True, "mode": "external", "auth_path": str(auth_path)}
+ except OSError as exc:
+ return {"connected": False, "mode": "", "auth_path": "", "error": str(exc)}
+
+ return {"connected": False, "mode": "", "auth_path": ""}
+
+
+def _env_file_has_key(path: Path) -> bool:
+ if not path.is_file():
+ return False
+ for line in path.read_text(encoding="utf-8", errors="ignore").splitlines():
+ raw = line.strip()
+ if not raw or raw.startswith("#") or "=" not in raw:
+ continue
+ key, value = raw.split("=", 1)
+ if key.strip() in _AUTH_ENV_VARS and value.strip().strip("'\""):
+ return True
+ return False
+
+
+def _auth_store_has_credentials(path: Path) -> bool:
+ if not path.is_file() or path.stat().st_size <= 0:
+ return False
+ try:
+ data = json.loads(path.read_text(encoding="utf-8"))
+ except ValueError:
+ return False
+ return _contains_secret(data)
+
+
+def _contains_secret(value: Any) -> bool:
+ if isinstance(value, dict):
+ for key, item in value.items():
+ if str(key) in _SECRET_KEYS and isinstance(item, str) and len(item.strip()) > 3:
+ return True
+ if _contains_secret(item):
+ return True
+ if isinstance(value, list):
+ return any(_contains_secret(item) for item in value)
+ return False
diff --git a/plugins/_orchestrator/helpers/adapters/opencode.py b/plugins/_orchestrator/helpers/adapters/opencode.py
new file mode 100644
index 000000000..ec140dee6
--- /dev/null
+++ b/plugins/_orchestrator/helpers/adapters/opencode.py
@@ -0,0 +1,74 @@
+from __future__ import annotations
+
+import json
+import os
+from pathlib import Path
+from typing import Any
+
+from plugins._orchestrator.helpers.adapters.base import TerminalAgentAdapter
+
+
+_ENV_KEYS = {
+ "ANTHROPIC_API_KEY",
+ "DEEPSEEK_API_KEY",
+ "GEMINI_API_KEY",
+ "GH_TOKEN",
+ "GITHUB_TOKEN",
+ "GOOGLE_API_KEY",
+ "GROQ_API_KEY",
+ "OPENAI_API_KEY",
+ "OPENROUTER_API_KEY",
+ "XAI_API_KEY",
+}
+
+_SECRET_KEYS = {"access_token", "api_key", "refresh_token", "token"}
+
+
+class OpenCodeAdapter(TerminalAgentAdapter):
+ id = "opencode"
+ title = "OpenCode"
+ binary = "opencode"
+ install_hint = "curl -fsSL https://opencode.ai/install | bash"
+ description = "OpenCode CLI in non-interactive run mode."
+
+ def auth_status(self, config: dict[str, Any] | None = None) -> dict[str, Any]:
+ env_var = next((name for name in sorted(_ENV_KEYS) if os.environ.get(name)), "")
+ if env_var:
+ return {"connected": True, "mode": "env", "auth_path": env_var}
+
+ path = _auth_path()
+ try:
+ if _auth_store_has_credentials(path):
+ return {"connected": True, "mode": "external", "auth_path": str(path)}
+ except OSError as exc:
+ return {"connected": False, "mode": "", "auth_path": "", "error": str(exc)}
+
+ return {"connected": False, "mode": "", "auth_path": str(path)}
+
+
+def _auth_path() -> Path:
+ data_home = os.environ.get("XDG_DATA_HOME", "").strip()
+ root = Path(data_home).expanduser() if data_home else Path.home() / ".local" / "share"
+ return root / "opencode" / "auth.json"
+
+
+def _auth_store_has_credentials(path: Path) -> bool:
+ if not path.is_file() or path.stat().st_size <= 0:
+ return False
+ try:
+ data = json.loads(path.read_text(encoding="utf-8"))
+ except ValueError:
+ return False
+ return _contains_secret(data)
+
+
+def _contains_secret(value: Any) -> bool:
+ if isinstance(value, dict):
+ for key, item in value.items():
+ if str(key) in _SECRET_KEYS and isinstance(item, str) and item.strip():
+ return True
+ if _contains_secret(item):
+ return True
+ if isinstance(value, list):
+ return any(_contains_secret(item) for item in value)
+ return False
diff --git a/plugins/_orchestrator/helpers/registry.py b/plugins/_orchestrator/helpers/registry.py
new file mode 100644
index 000000000..b13865f72
--- /dev/null
+++ b/plugins/_orchestrator/helpers/registry.py
@@ -0,0 +1,44 @@
+from __future__ import annotations
+
+from typing import Any
+
+from plugins._orchestrator.helpers.adapters.a0 import AgentZeroAdapter
+from plugins._orchestrator.helpers.adapters.base import TerminalAgentAdapter
+from plugins._orchestrator.helpers.adapters.claude import ClaudeCodeAdapter
+from plugins._orchestrator.helpers.adapters.codex import CodexAdapter
+from plugins._orchestrator.helpers.adapters.cursor import CursorCliAdapter
+from plugins._orchestrator.helpers.adapters.grok import GrokBuildAdapter
+from plugins._orchestrator.helpers.adapters.hermes import HermesAgentAdapter
+from plugins._orchestrator.helpers.adapters.opencode import OpenCodeAdapter
+
+# Register new terminal agent adapters here.
+_ADAPTERS: dict[str, TerminalAgentAdapter] = {
+ adapter.id: adapter
+ for adapter in (
+ AgentZeroAdapter(),
+ CodexAdapter(),
+ ClaudeCodeAdapter(),
+ CursorCliAdapter(),
+ GrokBuildAdapter(),
+ HermesAgentAdapter(),
+ OpenCodeAdapter(),
+ )
+}
+
+
+def get_adapter(agent_id: str) -> TerminalAgentAdapter:
+ adapter = _ADAPTERS.get(str(agent_id or "").strip().lower())
+ if adapter is None:
+ known = ", ".join(sorted(_ADAPTERS))
+ raise ValueError(f"Unknown terminal agent '{agent_id}'. Available: {known}")
+ return adapter
+
+
+def list_adapters() -> list[TerminalAgentAdapter]:
+ return list(_ADAPTERS.values())
+
+
+def adapter_config(agent_id: str, plugin_config: dict[str, Any] | None) -> dict[str, Any]:
+ source = plugin_config if isinstance(plugin_config, dict) else {}
+ raw = source.get(agent_id)
+ return raw if isinstance(raw, dict) else {}
diff --git a/plugins/_orchestrator/plugin.yaml b/plugins/_orchestrator/plugin.yaml
new file mode 100644
index 000000000..bece5adfd
--- /dev/null
+++ b/plugins/_orchestrator/plugin.yaml
@@ -0,0 +1,8 @@
+name: _orchestrator
+title: Orchestrator
+description: Load-on-demand skill and status UI for external headless terminal coding agents such as Agent Zero, Codex, Claude Code, Cursor CLI, Grok Build, Hermes Agent, and OpenCode.
+version: 0.1.0
+settings_sections:
+ - external
+per_project_config: false
+per_agent_config: false
diff --git a/plugins/_orchestrator/skills/AGENTS.md b/plugins/_orchestrator/skills/AGENTS.md
new file mode 100644
index 000000000..69fc91e3b
--- /dev/null
+++ b/plugins/_orchestrator/skills/AGENTS.md
@@ -0,0 +1,42 @@
+# Orchestrator Skills - AGENTS.md
+
+## Purpose
+
+- Own the plugin's load-on-demand skill collection.
+- Keep external-agent orchestration instructions available only when the user asks for terminal/headless agent delegation.
+
+## Ownership
+
+- Owns skill folders under `skills/`, currently `orchestrator/`.
+- Does not own Python adapter code, settings UI, or credential storage.
+
+## Local Contracts
+
+- Skills must use Agent Zero's ordinary shell/code execution tools, either container `code_execution_tool` or A0 Connector `code_execution_remote`, not a plugin-specific terminal agent tool.
+- Skill instructions plus their local reference files must be self-contained enough for the agent to run install checks, login/setup, smoke prompts, real tasks, and verification.
+- Generic orchestration belongs in `SKILL.md`; per-agent details belong in the nearest `references/` file.
+- Non-A0 agents first choose host-vs-container execution. Unknown preferences must be asked once and saved with memory when available.
+- Host-machine agents use A0 CLI remote execution; container agents use the human-in-the-loop setup loop.
+- A0 Headless remains the setup exception and must ask for same-instance vs another/spun-up instance when the target is ambiguous.
+- The skill must instruct agents not to paste or request secrets in chat when safer browser/device/login-prompt paths exist.
+- The skill must instruct agents not to use Computer Use to operate coding-agent terminals or TUIs.
+
+## Work Guidance
+
+- Keep the SKILL.md text operational, not encyclopedic.
+- Add or update per-agent reference files when CLI-specific commands or auth behavior change.
+- When CLI flags change, update the exact command blocks and the regression tests that assert important phrases.
+- If a loaded chat keeps old behavior, start a fresh chat or reload `orchestrator`.
+
+## Verification
+
+- Run the skill text contract test:
+ ```bash
+ docker exec 8dc967046cda bash -lc 'cd /a0 && /opt/venv-a0/bin/python plugins/_orchestrator/tests/test_status_adapters.py'
+ ```
+
+## Child DOX Index
+
+| Child | Scope |
+| --- | --- |
+| [orchestrator/AGENTS.md](orchestrator/AGENTS.md) | The `orchestrator` skill manifest and delegation workflow body. |
diff --git a/plugins/_orchestrator/skills/orchestrator/AGENTS.md b/plugins/_orchestrator/skills/orchestrator/AGENTS.md
new file mode 100644
index 000000000..ae0acba5b
--- /dev/null
+++ b/plugins/_orchestrator/skills/orchestrator/AGENTS.md
@@ -0,0 +1,48 @@
+# orchestrator Skill - AGENTS.md
+
+## Purpose
+
+- Teach Agent Zero how to delegate work to external terminal/headless coding agents only after the skill is loaded.
+- Keep generic orchestration rules in `SKILL.md` and agent-specific command runbooks in `references/`.
+- Provide copy-ready command patterns for A0 Headless, Codex CLI, Claude Code, Cursor CLI, Grok Build, Hermes Agent, and OpenCode through reference files.
+
+## Ownership
+
+- Owns `SKILL.md`.
+- Owns per-agent references under `references/`.
+- Does not own adapter status detection, UI state, or installation APIs.
+
+## Local Contracts
+
+- Front matter must allow `code_execution_tool`, `code_execution_remote`, `memory_load`, and `memory_save`.
+- The body must explicitly state that no `terminal_agent` tool exists.
+- The setup flow must ask or recall host-vs-container preference for each coding agent before running non-A0 adapters.
+- The host flow must use A0 CLI `code_execution_remote`; the container flow must check install, install only the requested CLI if missing, probe version/help, run a smoke prompt, handle auth/setup with the user, then retry before the real task.
+- Agent-specific commands, auth quirks, install hints, and smoke prompts belong in the matching `references/<agent>.md` file.
+- `SKILL.md` must tell agents to read only the relevant reference file before acting.
+- Claude Code reference guidance must avoid plain `claude` and bare `claude auth login` until the user chooses `--claudeai`, `--console`, `--sso`, or external API-key auth.
+- Claude Code API-key auth must source Agent Zero secrets from `/a0/usr/.env`, not from the workdir.
+- If a full-screen Claude Code TUI is already open, the Claude reference must reset the terminal session instead of sending Enter or `/login`.
+- Root shells must not use Claude Code `bypassPermissions`; non-root shells may use it as the default non-interactive mode.
+- Long-running commands should stay in a shell session and be polled there.
+- Computer Use must not be used to drive coding-agent terminals or TUIs.
+
+## Work Guidance
+
+- Put exact commands in fenced bash blocks when the agent needs to execute them.
+- State boundaries in imperative language so Agent Zero follows them during chat.
+- Keep expected smoke responses deterministic.
+- Keep `SKILL.md` short; put growing per-agent details in references.
+
+## Verification
+
+- Run:
+ ```bash
+ docker exec 8dc967046cda bash -lc 'cd /a0 && /opt/venv-a0/bin/python plugins/_orchestrator/tests/test_status_adapters.py'
+ ```
+
+## Child DOX Index
+
+| Child | Scope |
+| --- | --- |
+| [references/AGENTS.md](references/AGENTS.md) | Per-agent command, setup, auth, smoke, and real-task runbooks. |
diff --git a/plugins/_orchestrator/skills/orchestrator/SKILL.md b/plugins/_orchestrator/skills/orchestrator/SKILL.md
new file mode 100644
index 000000000..87ae72461
--- /dev/null
+++ b/plugins/_orchestrator/skills/orchestrator/SKILL.md
@@ -0,0 +1,105 @@
+---
+name: orchestrator
+description: Use when delegating coding or repository work to external terminal coding agents such as the user's host Claude Code/Codex/Cursor CLI or container-installed pal agents.
+triggers:
+ - "terminal agent"
+ - "external coding agent"
+ - "delegate to codex"
+ - "delegate to claude code"
+ - "delegate to cursor"
+ - "delegate to cursor cli"
+ - "delegate to grok build"
+ - "delegate to a0 headless"
+ - "delegate to hermes"
+ - "delegate to opencode"
+allowed_tools:
+ - code_execution_tool
+ - code_execution_remote
+ - memory_load
+ - memory_save
+---
+
+# Orchestrator
+
+Use shell/code execution to run external terminal agents. There is no `terminal_agent` tool: this skill exists so these heavy delegation instructions load only when the task needs them.
+
+Prefer the user's own host-machine CLI when that is what they mean. Container-installed agents are the pal scenario: useful and supported, but secondary.
+
+## Rules
+
+- For Codex, Claude Code, Cursor CLI, Grok Build, Hermes Agent, and OpenCode, first decide the execution place: the user's local machine through A0 CLI, or the Agent Zero Docker/container shell.
+- If the user did not specify local/host versus container, check memory for this coding agent's execution-place preference. If no current preference is known, ask: "Do you want me to use your own local <agent> through A0 CLI, or the <agent> installed inside the Agent Zero container?"
+- After the user chooses, save a stable per-agent preference with `memory_save`, for example: "For orchestrator, the user prefers Claude Code to run on the host machine through A0 CLI by default." If memory tools are unavailable, continue without saving.
+- Never use Computer Use to drive coding-agent terminals, menus, or TUIs. Use headless CLI commands through `code_execution_remote` or `code_execution_tool`; if that is not possible, stop and ask.
+- ACP may be available as a community plugin, but do not assume it is installed. Mention it only if the user explicitly asks for ACP or the direct CLI path is unsuitable.
+- For Codex, Claude Code, Cursor CLI, Grok Build, Hermes Agent, and OpenCode, setup is part of the workflow: check whether the CLI is installed, install only the requested CLI if missing, probe its version/help, run a tiny smoke prompt, then run the real task.
+- Keep authentication human-in-the-loop. You may start the CLI login/setup command, relay the exact URL, device code, browser step, or prompt to the user, then wait for the user to confirm completion before retrying the smoke prompt.
+- Never start a full-screen CLI/TUI as a login fallback. If you accidentally opened one and see welcome, theme, provider, or unreadable menu output, reset the terminal session instead of sending keys into it.
+- If login/setup shows a menu or provider choices, show those choices to the user in chat and ask which one to select. Keep the terminal session open, then send the user's selected number/key back to that session.
+- Never ask the user to paste secrets into chat unless there is no safer path. Prefer browser/device login, the CLI's own prompt, or an environment variable set outside chat.
+- Always choose an explicit working directory for repository work. Prefer the user's project path over Agent Zero's default workdir.
+- Optional command defaults may be stored in `/a0/usr/plugins/_orchestrator/config.json`. Read only the adapter block you need, and never print secrets.
+- For long-running commands, start the CLI in a shell session and poll that session's output. Do not add your own timeout wrapper around the terminal agent.
+- Pass a self-contained task brief: goal, target files or repo path, constraints, verification commands, and expected output.
+- After the terminal agent finishes, inspect its output and verify important changes yourself before reporting success.
+
+## Host CLI Flow
+
+Use this when the user wants their own local Claude Code, Codex, Cursor CLI, Grok Build, Hermes Agent, or OpenCode.
+
+1. Use `code_execution_remote`, not `code_execution_tool`, because paths, shell, login, and installed CLIs belong to the A0 CLI host machine.
+2. If `code_execution_remote` is unavailable or reports no connected CLI / remote execution disabled, tell the user exactly:
+
+```text
+To use your local coding agents, open a terminal on your computer and install A0 CLI if needed.
+
+macOS / Linux:
+curl -LsSf https://cli.agent-zero.ai/install.sh | sh
+
+Windows PowerShell:
+irm https://cli.agent-zero.ai/install.ps1 | iex
+
+If A0 CLI is already installed but not running, open a terminal and run:
+a0
+
+Connect it to this Agent Zero instance. Select the chat/session, or type the Agent Zero URL manually if this is a remote/VPS instance. In A0 CLI, press F4 to allow Remote Code Execution. Press F3 too if the task needs host file writes. Then ask me again to use your local <agent>.
+```
+
+3. Once remote execution is available, optionally load `host-code-execution` for host-shell safety rules.
+4. Read the relevant agent reference file. Run the same install/probe, smoke, auth, and real-task commands with `code_execution_remote`.
+5. Keep workdir semantics host-side: `cd` to the host project path, not `/a0/usr/workdir`, unless the user explicitly wants the container workspace.
+
+## Container Pal Flow
+
+Use this when the user chooses the Agent Zero container or explicitly wants a pal agent inside Docker.
+
+For Agent Zero Headless, read `references/a0.md` and follow its target-selection flow instead of the non-A0 setup loop.
+
+For every non-A0 container adapter:
+
+1. Read only the relevant reference file:
+
+```bash
+# Use skills_tool action=read_file on the matching file under references/.
+```
+
+2. Check install with the reference command.
+3. If missing, install only that requested CLI with the command listed in its reference, then run the install check again.
+4. Probe `--version` or `--help`.
+5. Run the reference smoke prompt.
+6. If the smoke prompt reports missing auth, run only the login/setup command named in that reference; do not run the bare CLI as a login fallback. Relay URLs, device codes, prompts, and visible menu choices to the user in chat, then wait. If a prior command left a TUI open, reset the terminal session before continuing.
+7. After the user confirms, rerun the smoke prompt. Only then run the real task.
+
+You can consult both host and container agents in one workflow. Keep their shell sessions separate, label which agent/location produced each answer, and compare results before acting.
+
+## Reference Files
+
+- Agent Zero Headless: `references/a0.md`
+- Codex CLI: `references/codex.md`
+- Claude Code: `references/claude.md`
+- Cursor CLI: `references/cursor.md`
+- Grok Build: `references/grok.md`
+- Hermes Agent: `references/hermes.md`
+- OpenCode: `references/opencode.md`
+
+Keep generic orchestration rules here. Put agent-specific commands, auth quirks, install hints, and smoke prompts in the matching reference file.
diff --git a/plugins/_orchestrator/skills/orchestrator/references/AGENTS.md b/plugins/_orchestrator/skills/orchestrator/references/AGENTS.md
new file mode 100644
index 000000000..6825f7bbe
--- /dev/null
+++ b/plugins/_orchestrator/skills/orchestrator/references/AGENTS.md
@@ -0,0 +1,34 @@
+# Terminal Agent References - AGENTS.md
+
+## Purpose
+
+- Own per-agent command references for the `orchestrator` skill.
+- Keep `SKILL.md` generic while preserving exact install, auth, smoke, and run commands for each supported terminal agent.
+
+## Ownership
+
+- Owns one markdown file per agent id, currently `a0.md`, `codex.md`, `claude.md`, `cursor.md`, `grok.md`, `hermes.md`, and `opencode.md`.
+- Does not own adapter status code, settings UI, or global orchestration rules.
+
+## Local Contracts
+
+- Each reference must include when to use the agent, install/probe commands, auth/setup behavior, smoke command, and real-task command shape.
+- Generic rules such as no bare CLI fallback, no secrets in chat, explicit workdir, and smoke-before-real stay in parent `SKILL.md`.
+- Reference filenames must match adapter ids where possible.
+- When adding a reference, update the parent `SKILL.md` reference index, README adding-agent steps, and tests.
+
+## Work Guidance
+
+- Keep each file operational and short. The agent should be able to copy the command block safely.
+- Redact or avoid secrets; use environment variables and Agent Zero secret paths without printing values.
+
+## Verification
+
+- Run:
+ ```bash
+ docker exec 8dc967046cda bash -lc 'cd /a0 && /opt/venv-a0/bin/python plugins/_orchestrator/tests/test_status_adapters.py'
+ ```
+
+## Child DOX Index
+
+This folder has no child DOX files.
diff --git a/plugins/_orchestrator/skills/orchestrator/references/a0.md b/plugins/_orchestrator/skills/orchestrator/references/a0.md
new file mode 100644
index 000000000..89b080ad6
--- /dev/null
+++ b/plugins/_orchestrator/skills/orchestrator/references/a0.md
@@ -0,0 +1,31 @@
+# Agent Zero Headless
+
+Use `a0 headless` when the user wants another Agent Zero runtime to help, or as a secondary use case when this same Agent Zero instance should ask itself a focused question through the local headless interface.
+
+A0 is the setup exception. Do not run a generic login flow. If the user did not specify the target, ask whether they want to use this same Agent Zero instance or another/spun-up instance. For another instance, ask for or use a configured host. For protected hosts, use configured shell environment credentials when present.
+
+Avoid recursive loops. Tell the target Agent Zero instance to answer directly and not to delegate back through terminal agents.
+
+## Install And Probe
+
+```bash
+command -v a0 >/dev/null || test -x /opt/venv/bin/a0
+a0 --help >/dev/null || /opt/venv/bin/a0 --help >/dev/null
+```
+
+## New Chat
+
+```bash
+cd "$WORKDIR"
+a0 headless --host http://localhost:80 --no-docker-discovery --output jsonl --new-chat --workspace "$WORKDIR" -p "$TASK"
+```
+
+Use a configured remote host instead of `http://localhost:80` when the settings screen or user provides one.
+
+## Continue Chat
+
+```bash
+a0 headless --host "$A0_HOST" --no-docker-discovery --output jsonl --chat "$CONTEXT_ID" --workspace "$WORKDIR" -p "$TASK"
+```
+
+For protected hosts, use the CLI's supported environment variables, for example `A0_USERNAME` and `A0_PASSWORD`, rather than inventing username/password flags.
diff --git a/plugins/_orchestrator/skills/orchestrator/references/claude.md b/plugins/_orchestrator/skills/orchestrator/references/claude.md
new file mode 100644
index 000000000..82937e54b
--- /dev/null
+++ b/plugins/_orchestrator/skills/orchestrator/references/claude.md
@@ -0,0 +1,50 @@
+# Claude Code
+
+Use Claude Code in print mode. Skip permissions is the default non-interactive mode only when the shell is not root; Claude Code rejects that mode under root/sudo.
+
+## Install And Probe
+
+```bash
+command -v claude >/dev/null || npm install -g @anthropic-ai/claude-code
+claude --version
+claude auth status || true
+```
+
+## Smoke Prompt
+
+```bash
+cd "$WORKDIR"
+if [ "$(id -u)" -eq 0 ]; then
+ claude -p "Respond exactly: TERMINAL_AGENT_SMOKE_OK" --output-format json
+else
+ claude -p "Respond exactly: TERMINAL_AGENT_SMOKE_OK" --output-format json --permission-mode bypassPermissions --allowedTools Bash,Read,Edit
+fi
+```
+
+## Login
+
+If Claude Code reports `Not logged in` or asks for `/login`, do not open the TUI. First ask the user which auth mode to use:
+
+1. Claude subscription: `claude auth login --claudeai`
+2. Anthropic Console/API billing: `claude auth login --console`
+3. SSO: `claude auth login --sso`
+4. API key outside chat: ask the user to open Settings > External Services > Secrets Management and enter the key in `/a0/usr/.env`, right after `ANTHROPIC_API_KEY=`. The Agent Zero `.env` file is `/a0/usr/.env`, not `$WORKDIR/.env`.
+
+After the user chooses, run only that command and relay its browser/device instructions. If the user gives an email to prefill, add `--email "<email>"`. Wait for confirmation, then retry the smoke prompt. Do not redirect them to a Docker shell just to choose a menu option. Do not start plain `claude` for login; it opens the first-run TUI (theme/provider menus) and is not a safe agent-driven login surface. Do not run bare `claude auth login` when provider choice is still unknown.
+
+If a previous attempt already opened plain `claude` and the terminal shows theme/provider menus or no readable login URL, stop. Reset that terminal session; do not press Enter, do not send `/login`, and do not keep polling the stuck TUI. Then ask for the auth mode above and run the matching `claude auth login ...` command in a fresh terminal session.
+
+## Agent Zero Secrets
+
+When using an API key from Agent Zero secrets, source `/a0/usr/.env` without printing it. Some Agent Zero installs store the Anthropic key as `API_KEY_ANTHROPIC`; map it to the CLI variable before running Claude Code:
+
+```bash
+set -a
+. /a0/usr/.env
+set +a
+[ -z "${ANTHROPIC_API_KEY:-}" ] && [ -n "${API_KEY_ANTHROPIC:-}" ] && export ANTHROPIC_API_KEY="$API_KEY_ANTHROPIC"
+```
+
+## Real Task
+
+Replace the smoke prompt with `$TASK`.
diff --git a/plugins/_orchestrator/skills/orchestrator/references/codex.md b/plugins/_orchestrator/skills/orchestrator/references/codex.md
new file mode 100644
index 000000000..6bec3cffa
--- /dev/null
+++ b/plugins/_orchestrator/skills/orchestrator/references/codex.md
@@ -0,0 +1,36 @@
+# Codex CLI
+
+Use Codex for autonomous coding tasks in a repository. Inside the Agent Zero Docker runtime, the Codex sandbox may fail, so the Docker-safe default uses `--dangerously-bypass-approvals-and-sandbox`.
+
+## Install And Probe
+
+```bash
+command -v codex >/dev/null || npm install -g @openai/codex
+codex --version
+```
+
+## Optional Plugin-Owned Login
+
+If the settings screen connected Codex with plugin-owned device login, run Codex with:
+
+```bash
+export CODEX_HOME=/a0/usr/plugins/_orchestrator/data/codex
+```
+
+If Codex reports missing auth, prefer the plugin settings device-code login when available. Otherwise run `codex login`, relay the device/browser instructions to the user, wait for confirmation, and retry the smoke prompt.
+
+## Smoke Prompt
+
+```bash
+cd "$WORKDIR"
+codex exec --skip-git-repo-check --dangerously-bypass-approvals-and-sandbox "Respond exactly: TERMINAL_AGENT_SMOKE_OK"
+```
+
+## Real Task
+
+```bash
+cd "$WORKDIR"
+codex exec --skip-git-repo-check --dangerously-bypass-approvals-and-sandbox "$TASK"
+```
+
+If the runtime supports Codex's own sandbox, omit `--dangerously-bypass-approvals-and-sandbox`. If the settings screen has a model override, add `-m "$MODEL"`.
diff --git a/plugins/_orchestrator/skills/orchestrator/references/cursor.md b/plugins/_orchestrator/skills/orchestrator/references/cursor.md
new file mode 100644
index 000000000..9b07ba5da
--- /dev/null
+++ b/plugins/_orchestrator/skills/orchestrator/references/cursor.md
@@ -0,0 +1,54 @@
+# Cursor CLI
+
+Use Cursor CLI for headless Cursor Agent tasks. The official binary is `agent`; running it without `-p` starts the interactive terminal UI, so use `agent -p` for Agent Zero delegation.
+
+## Install And Probe
+
+```bash
+command -v agent >/dev/null || curl https://cursor.com/install -fsS | bash
+agent --version
+agent status || true
+```
+
+## Smoke Prompt
+
+```bash
+cd "$WORKDIR"
+agent -p --output-format text "Respond exactly: TERMINAL_AGENT_SMOKE_OK"
+```
+
+## Login
+
+For scripts and containers, prefer API-key auth:
+
+```bash
+export CURSOR_API_KEY="..."
+```
+
+Do not ask the user to paste the key into chat. Ask them to set `CURSOR_API_KEY` in the runtime environment, or to add it through Settings > External Services > Secrets Management when that maps into `/a0/usr/.env`.
+
+When using Agent Zero secrets, source `/a0/usr/.env` without printing it. If the key is stored as `API_KEY_CURSOR`, map it to the CLI variable before running Cursor CLI:
+
+```bash
+set -a
+. /a0/usr/.env
+set +a
+[ -z "${CURSOR_API_KEY:-}" ] && [ -n "${API_KEY_CURSOR:-}" ] && export CURSOR_API_KEY="$API_KEY_CURSOR"
+```
+
+For browser login in a container or remote terminal, run:
+
+```bash
+NO_OPEN_BROWSER=1 agent login
+```
+
+Relay the printed URL to the user, wait for confirmation, then retry the smoke prompt. Use `agent status` to check auth and `agent logout` only when the user explicitly asks to clear Cursor CLI credentials.
+
+## Real Task
+
+```bash
+cd "$WORKDIR"
+agent -p --force --output-format text "$TASK"
+```
+
+Use `--output-format json` when the caller needs structured output, or `--output-format stream-json --stream-partial-output` when polling progress. Select models with Cursor CLI's `/model` command in an interactive setup, not with an invented flag.
diff --git a/plugins/_orchestrator/skills/orchestrator/references/grok.md b/plugins/_orchestrator/skills/orchestrator/references/grok.md
new file mode 100644
index 000000000..5280ab494
--- /dev/null
+++ b/plugins/_orchestrator/skills/orchestrator/references/grok.md
@@ -0,0 +1,57 @@
+# Grok Build
+
+Use Grok Build for headless xAI coding-agent tasks. Running `grok` with no arguments starts the interactive TUI, so use `-p` for Agent Zero delegation.
+
+## Install And Probe
+
+```bash
+command -v grok >/dev/null || curl -fsSL https://x.ai/cli/install.sh | bash
+grok version || grok --version || grok --help
+```
+
+If the installer host is blocked, use the official npm distribution instead:
+
+```bash
+npm install -g @xai-official/grok
+```
+
+## Smoke Prompt
+
+```bash
+grok --no-auto-update --cwd "$WORKDIR" -p "Respond exactly: TERMINAL_AGENT_SMOKE_OK" --output-format json --always-approve --no-alt-screen
+```
+
+## Login
+
+If Grok Build needs auth, prefer API-key auth for headless automation:
+
+```bash
+export XAI_API_KEY="xai-..."
+```
+
+Do not ask the user to paste the key into chat. Ask them to set `XAI_API_KEY` in the runtime environment, or to add it through Settings > External Services > Secrets Management when that maps into `/a0/usr/.env`.
+
+When using Agent Zero secrets, source `/a0/usr/.env` without printing it. Some Agent Zero installs store the xAI key as `API_KEY_XAI`; map it to the CLI variable before running Grok Build:
+
+```bash
+set -a
+. /a0/usr/.env
+set +a
+[ -z "${XAI_API_KEY:-}" ] && [ -n "${API_KEY_XAI:-}" ] && export XAI_API_KEY="$API_KEY_XAI"
+```
+
+For account login in a container or remote terminal, run:
+
+```bash
+grok login --device-auth
+```
+
+Relay the printed URL and user code to the user, wait for confirmation, then retry the smoke prompt. Do not run bare `grok`; it opens the full-screen TUI.
+
+## Real Task
+
+```bash
+grok --no-auto-update --cwd "$WORKDIR" -p "$TASK" --output-format json --always-approve --no-alt-screen
+```
+
+Add `-m "$MODEL"` only when the user or settings provide a model override. For session continuation, use `--session-id`, `--resume`, or `--continue` only when the user asks for it.
diff --git a/plugins/_orchestrator/skills/orchestrator/references/hermes.md b/plugins/_orchestrator/skills/orchestrator/references/hermes.md
new file mode 100644
index 000000000..4e210c95d
--- /dev/null
+++ b/plugins/_orchestrator/skills/orchestrator/references/hermes.md
@@ -0,0 +1,25 @@
+# Hermes Agent
+
+Use Hermes Agent for quiet headless chat. Skip approvals by default for non-interactive delegation.
+
+## Install And Probe
+
+```bash
+command -v hermes >/dev/null || curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
+hermes --version || hermes --help
+```
+
+## Smoke Prompt
+
+```bash
+cd "$WORKDIR"
+hermes chat --quiet --source tool -q "Respond exactly: TERMINAL_AGENT_SMOKE_OK" --yolo
+```
+
+## Login
+
+If Hermes needs setup, run `hermes setup --portal` when available, relay the OAuth/browser instructions to the user, wait for confirmation, and retry the smoke prompt. If it asks for a provider API key, have the user type it into the CLI prompt or set it as an environment variable outside chat.
+
+## Real Task
+
+Replace the smoke prompt with `$TASK`. Add `--model`, `--provider`, or `--toolsets` only when the user or settings provide them.
diff --git a/plugins/_orchestrator/skills/orchestrator/references/opencode.md b/plugins/_orchestrator/skills/orchestrator/references/opencode.md
new file mode 100644
index 000000000..867080916
--- /dev/null
+++ b/plugins/_orchestrator/skills/orchestrator/references/opencode.md
@@ -0,0 +1,24 @@
+# OpenCode
+
+OpenCode has a non-interactive run mode suitable for this skill.
+
+## Install And Probe
+
+```bash
+command -v opencode >/dev/null || curl -fsSL https://opencode.ai/install | bash
+opencode --version || opencode --help
+```
+
+## Smoke Prompt
+
+```bash
+opencode run --dir "$WORKDIR" --auto "Respond exactly: TERMINAL_AGENT_SMOKE_OK"
+```
+
+## Login
+
+If OpenCode needs auth, run `opencode auth login`, relay the provider/browser/key prompt to the user, wait for confirmation, and retry the smoke prompt. If that command is unavailable, run `opencode`, use `/connect`, and keep the same human-in-the-loop boundary.
+
+## Real Task
+
+Replace the smoke prompt with `$TASK`. Add `--model` or `--agent` only when the user or settings provide them.
diff --git a/plugins/_orchestrator/tests/AGENTS.md b/plugins/_orchestrator/tests/AGENTS.md
new file mode 100644
index 000000000..578db4659
--- /dev/null
+++ b/plugins/_orchestrator/tests/AGENTS.md
@@ -0,0 +1,34 @@
+# Orchestrator Tests - AGENTS.md
+
+## Purpose
+
+- Provide lightweight regression checks for the plugin's source contracts.
+- Catch accidental reintroduction of the old tool/runner architecture and important skill instruction drift.
+
+## Ownership
+
+- Owns direct test scripts under `tests/`, currently `test_status_adapters.py`.
+- Does not own generated caches or live CLI credentials.
+
+## Local Contracts
+
+- Tests must run with the Agent Zero framework runtime, not the agent execution runtime.
+- Keep tests self-contained and deterministic. They may inspect files and adapter metadata but should not call network login endpoints or run real terminal agents.
+- When adding an adapter, assert registry order and status-only behavior as needed.
+- When changing setup/login guidance, add or update assertions for the phrases that protect the workflow.
+
+## Work Guidance
+
+- Prefer simple assert-based checks that can run as a script inside Docker.
+- Avoid broad snapshots of full docs; assert the contract phrases that matter.
+
+## Verification
+
+- Main check:
+ ```bash
+ docker exec 8dc967046cda bash -lc 'cd /a0 && /opt/venv-a0/bin/python plugins/_orchestrator/tests/test_status_adapters.py'
+ ```
+
+## Child DOX Index
+
+This folder has no child DOX files.
diff --git a/plugins/_orchestrator/tests/test_status_adapters.py b/plugins/_orchestrator/tests/test_status_adapters.py
new file mode 100644
index 000000000..7424626f2
--- /dev/null
+++ b/plugins/_orchestrator/tests/test_status_adapters.py
@@ -0,0 +1,230 @@
+import os
+from pathlib import Path
+import sys
+import tempfile
+
+repo_root = next(
+ parent for parent in Path(__file__).resolve().parents if (parent / "agent.py").is_file()
+)
+sys.path.insert(0, str(repo_root))
+
+from plugins._orchestrator.helpers.adapters.cursor import CursorCliAdapter
+from plugins._orchestrator.helpers.adapters.grok import _toml_has_secret
+from plugins._orchestrator.helpers.adapters.grok import GrokBuildAdapter
+from plugins._orchestrator.helpers.adapters.base import TerminalAgentAdapter
+from plugins._orchestrator.helpers.registry import list_adapters
+
+
+class DummyAdapter(TerminalAgentAdapter):
+ id = "dummy"
+ binary = "missing-dummy-agent"
+
+ def auth_status(self, config=None):
+ return {"connected": True, "mode": "external", "auth_path": ""}
+
+
+def test_configured_absolute_binary_is_installed():
+ with tempfile.TemporaryDirectory() as tmp:
+ binary = Path(tmp) / "agent"
+ binary.write_text("#!/bin/sh\n")
+ binary.chmod(0o755)
+
+ adapter = DummyAdapter()
+ config = {"binary": str(binary)}
+ assert adapter.resolve_binary(config) == str(binary)
+ assert adapter.is_installed(config)
+
+
+def test_registry_order_puts_a0_first():
+ assert [adapter.id for adapter in list_adapters()] == [
+ "a0",
+ "codex",
+ "claude",
+ "cursor",
+ "grok",
+ "hermes",
+ "opencode",
+ ]
+
+
+def test_adapters_are_status_only():
+ for adapter in list_adapters():
+ assert not hasattr(adapter, "install_command")
+ assert not hasattr(adapter, "build_command")
+ assert not hasattr(adapter, "parse_session_id")
+ assert not hasattr(adapter, "format_output")
+
+
+def test_tool_runner_files_are_removed():
+ plugin_root = Path(__file__).resolve().parents[1]
+ assert not (plugin_root / "tools" / "terminal_agent.py").exists()
+ assert not (plugin_root / "helpers" / "runner.py").exists()
+ assert not (
+ plugin_root / "prompts" / "agent.system.tool.terminal_agent.md"
+ ).exists()
+
+
+def test_claude_defaults_skip_permissions():
+ config_text = (Path(__file__).resolve().parents[1] / "default_config.yaml").read_text()
+ assert "permission_mode: bypassPermissions" in config_text
+
+
+def test_grok_defaults_headless_automation():
+ config_text = (Path(__file__).resolve().parents[1] / "default_config.yaml").read_text()
+ assert "grok:" in config_text
+ assert "output_format: json" in config_text
+ assert "always_approve: true" in config_text
+ assert "no_auto_update: true" in config_text
+
+
+def test_cursor_defaults_headless_automation():
+ config_text = (Path(__file__).resolve().parents[1] / "default_config.yaml").read_text()
+ assert "cursor:" in config_text
+ assert "binary: agent" in config_text
+ assert "output_format: text" in config_text
+ assert "force: true" in config_text
+
+
+def test_cursor_detects_agent_zero_cursor_env_key():
+ old_cursor = os.environ.pop("CURSOR_API_KEY", None)
+ old_a0_cursor = os.environ.get("API_KEY_CURSOR")
+ try:
+ os.environ["API_KEY_CURSOR"] = "secret"
+ assert CursorCliAdapter().auth_status()["auth_path"] == "API_KEY_CURSOR"
+ finally:
+ if old_cursor is not None:
+ os.environ["CURSOR_API_KEY"] = old_cursor
+ if old_a0_cursor is None:
+ os.environ.pop("API_KEY_CURSOR", None)
+ else:
+ os.environ["API_KEY_CURSOR"] = old_a0_cursor
+
+
+def test_grok_env_key_requires_present_environment_value():
+ old_value = os.environ.pop("GROK_TEST_KEY", None)
+ try:
+ assert not _toml_has_secret('env_key = "GROK_TEST_KEY"')
+ os.environ["GROK_TEST_KEY"] = "secret"
+ assert _toml_has_secret('env_key = "GROK_TEST_KEY"')
+ finally:
+ if old_value is None:
+ os.environ.pop("GROK_TEST_KEY", None)
+ else:
+ os.environ["GROK_TEST_KEY"] = old_value
+
+
+def test_grok_detects_agent_zero_xai_env_key():
+ old_xai = os.environ.pop("XAI_API_KEY", None)
+ old_a0_xai = os.environ.get("API_KEY_XAI")
+ try:
+ os.environ["API_KEY_XAI"] = "secret"
+ assert GrokBuildAdapter().auth_status()["auth_path"] == "API_KEY_XAI"
+ finally:
+ if old_xai is not None:
+ os.environ["XAI_API_KEY"] = old_xai
+ if old_a0_xai is None:
+ os.environ.pop("API_KEY_XAI", None)
+ else:
+ os.environ["API_KEY_XAI"] = old_a0_xai
+
+
+def test_skill_documents_human_setup_loop_and_a0_exception():
+ skill_root = Path(__file__).resolve().parents[1] / "skills" / "orchestrator"
+ skill_text = (skill_root / "SKILL.md").read_text()
+ references = skill_root / "references"
+ a0_text = (references / "a0.md").read_text()
+ codex_text = (references / "codex.md").read_text()
+ claude_text = (references / "claude.md").read_text()
+ cursor_text = (references / "cursor.md").read_text()
+ grok_text = (references / "grok.md").read_text()
+ hermes_text = (references / "hermes.md").read_text()
+ opencode_text = (references / "opencode.md").read_text()
+
+ assert "setup is part of the workflow" in skill_text
+ assert "code_execution_remote" in skill_text
+ assert "memory_load" in skill_text
+ assert "memory_save" in skill_text
+ assert "own local <agent>" in skill_text
+ assert "Agent Zero Docker/container shell" in skill_text
+ assert "For orchestrator, the user prefers Claude Code" in skill_text
+ assert "Never use Computer Use" in skill_text
+ assert "ACP may be available as a community plugin" in skill_text
+ assert "## Host CLI Flow" in skill_text
+ assert "curl -LsSf https://cli.agent-zero.ai/install.sh | sh" in skill_text
+ assert "irm https://cli.agent-zero.ai/install.ps1 | iex" in skill_text
+ assert "open a terminal and run:" in skill_text
+ assert "press F4 to allow Remote Code Execution" in skill_text
+ assert "Press F3 too if the task needs host file writes" in skill_text
+ assert "host-code-execution" in skill_text
+ assert "not `/a0/usr/workdir`" in skill_text
+ assert "## Container Pal Flow" in skill_text
+ assert "Keep authentication human-in-the-loop" in skill_text
+ assert "show those choices to the user in chat" in skill_text
+ assert "do not run the bare CLI as a login fallback" in skill_text
+ assert "read `references/a0.md`" in skill_text
+ assert "target-selection flow" in skill_text
+ assert "## Reference Files" in skill_text
+ assert "references/a0.md" in skill_text
+ assert "references/codex.md" in skill_text
+ assert "references/claude.md" in skill_text
+ assert "references/cursor.md" in skill_text
+ assert "references/grok.md" in skill_text
+ assert "references/hermes.md" in skill_text
+ assert "references/opencode.md" in skill_text
+
+ assert "A0 is the setup exception" in a0_text
+ assert "--new-chat" in a0_text
+ assert "--chat \"$CONTEXT_ID\"" in a0_text
+
+ assert "CODEX_HOME" in codex_text
+ assert "codex login" in codex_text
+ assert "--dangerously-bypass-approvals-and-sandbox" in codex_text
+
+ assert "claude auth login" in claude_text
+ assert "Do not start plain `claude`" in claude_text
+ assert "Never start a full-screen CLI/TUI" in skill_text
+ assert "Reset that terminal session" in claude_text
+ assert "do not press Enter, do not send `/login`" in claude_text
+ assert "Do not redirect them to a Docker shell just to choose a menu option" in claude_text
+ assert "claude auth login --claudeai" in claude_text
+ assert "claude auth login --console" in claude_text
+ assert "claude auth login --sso" in claude_text
+ assert "Settings > External Services > Secrets Management" in claude_text
+ assert "/a0/usr/.env" in claude_text
+ assert "ANTHROPIC_API_KEY" in claude_text
+ assert "API_KEY_ANTHROPIC" in claude_text
+ assert "Do not run bare `claude auth login`" in claude_text
+
+ assert "agent -p --output-format text" in cursor_text
+ assert "agent -p --force --output-format text" in cursor_text
+ assert "CURSOR_API_KEY" in cursor_text
+ assert "API_KEY_CURSOR" in cursor_text
+ assert "NO_OPEN_BROWSER=1 agent login" in cursor_text
+ assert "agent status" in cursor_text
+ assert "not with an invented flag" in cursor_text
+
+ assert "grok --no-auto-update --cwd \"$WORKDIR\" -p" in grok_text
+ assert "--output-format json" in grok_text
+ assert "--always-approve" in grok_text
+ assert "XAI_API_KEY" in grok_text
+ assert "API_KEY_XAI" in grok_text
+ assert "/a0/usr/.env" in grok_text
+ assert "grok login --device-auth" in grok_text
+ assert "Do not run bare `grok`" in grok_text
+
+ assert "hermes setup --portal" in hermes_text
+ assert "opencode auth login" in opencode_text
+
+
+if __name__ == "__main__":
+ test_configured_absolute_binary_is_installed()
+ test_registry_order_puts_a0_first()
+ test_adapters_are_status_only()
+ test_tool_runner_files_are_removed()
+ test_claude_defaults_skip_permissions()
+ test_cursor_defaults_headless_automation()
+ test_cursor_detects_agent_zero_cursor_env_key()
+ test_grok_defaults_headless_automation()
+ test_grok_env_key_requires_present_environment_value()
+ test_grok_detects_agent_zero_xai_env_key()
+ test_skill_documents_human_setup_loop_and_a0_exception()
diff --git a/plugins/_orchestrator/thumbnail.png b/plugins/_orchestrator/thumbnail.png
new file mode 100644
index 0000000000000000000000000000000000000000..2647a1bf120c860c9ff047d40ec8dc1969115837
GIT binary patch
literal 16585
zcmV)6K*+y|P)<h;3K|Lk000e1NJLTq00961009650{{R3u3WWM0000IP)t-sFG*Po
z5Ely%7YYv-2oDwo2n}&h!SetBKypb$K~#9!o&EoxB*l^E3BS#&=*nb7hD8e0)qR>s
z&lv4JD}hkYO0zqA63|-EYHoM-1UtiDsnx>X>iKCT(IauU&z>}*^<`1d%_!0{VCZT<
z85Rk2MfDfA=ZDDZZ-77>uUEe)x+}vi-ObI-KK3zld6AEMI_P#krU*@a`g0DXm_t*e
zmvij`G<6e*HrsO`MRCNT>>!IErGD%v`yVrSM~VQbAr7Rl`={9Tk(dJ!EP{(Tfm$3>
z2mQISDRI<(7eKnehQNjJ(L(G!{^7s#-~Rk+Sxx|KIshp)Mad*7B+ZjDDMAjA6a>h<
zDd-tNhoPJ@1sD)W<-ticoL3$G>Gd?e^AiQw{NMAklw;5nCfEd2C+l*80*WJ=1zb;v
z!zh54Y7+)QCIf&DHW2{tXYDX~+lqd2FxUSf<%dQKANu+~NV~-rD}auS8PY&%8fe;(
z2|1_GVV09R5EA1AyFruPvf=<z8CkRvBV_9QBl{!)@(+WO!ND;=G7eM`av2jLTARHM
zk+jkrxD2qW$O*tnfDj;$FSJ<=5U~J7{F(iJ0rHRPQFH*o1wvRWvZhH<T70oF$70iX
z?F3>bJBzjnkP#tLw;37VtfUC0iG%@L>Yt_Grvv$t%b_4iA(tltXcp6?rIfSwb8<Z#
zTr1A6k9!Z+QKmTbD(-D8q(ri0!6q~MQG+J~rA#+|wQeZmAmBc~z15GM0lyS$B{yUs
zn7cr3iNs99wrXc{O}h0cvlxpQ<bn~ItwT(K-=-W0<V_AJ<PIg1L2a4m3rqkBw72YI
zXW)lRNV;A&QId>XCI@XwkUZy1Do^r(o}3i8ioe~*Gy$3$C!5l2HrKU6ZkMaR*DNxH
z91v`>?Y4))9h<g3C_hGk&A9<lG|vT{B-L?~mEbFtROrd(MiHh(vXgD!xfkVH^Llxv
zt5Fr2WtQB=CPdvWVyl~p$&?dxwHs6?x!23fk6Z#b^Xv&AgN%kSLk5q5NKm1rNii&-
zW*%4&A_-DTP<J8YHpj>*8U&cH%UNa-ZGoYH5h8z_K2m^JKmw5*kdzz@Qi6~Or~9iS
z>yq5&Otf*|fi$3y%Z9Kt)sC=0X^2Lo-DY71ib28>gj*lA1h)ho03@&bXqps33jt-d
zE!oVQ(n^)-=I!EGPf9y0q|Me0uld}`k-C&FSk_)yt4&JxDWGo+CE}wJC=*5W#xLTB
zXW%Db2@(Vrxv~fjEo96qa!sA_l#}HUkd?O$G0duJQLi>=)dcrMTIec6WCmRz;}+C1
zAp>c(TWkOPhnL{puQ)_fgd8QHnGD2~z#=JG@^Haph#HtFK6kR`k|arE3!1z2vY81%
zo16%l(pCXx)m(Q2Wk};%ZeMwD3H<F70A>ORsSreftbNHWc)vX34>#9jVsAY>oZ0%h
z&f;8FOK%r#JdiTC=Q{P_+}j(@U0aG`CzJK(tVC$LzUvj6xNf!RCkjkvVRHWr{H}0-
z+(KUAF=Qd*iB*Jc3K8Z_%$jC#7E3}C3MbdzMu`ORP&H~cU+3DWfP_<pKtih0qb3vy
zF|Bmm0pYa=X5hb1b3bvBnb@>!!o;_4^`)=ldYLp&XOgU>NMUC4RWHNwahR2?OT5I9
z&B<ilOMhIi7W1@3Rwc7*=kWfpS>o*gGMTvKHD}JFNui(2=J(COtwj+u2?R18EU#vD
zWEE3P5C{-SZCoT-)jI2zlIPyeAVBCTXBK0o?v`yL=e$fFMxC$89$5;?noy}Gyvmt^
zaOpk)e!?9CJx3yK9Z1wGs78YxGYdi=qnM*fX~^@Y*&Umy17g!BkFiY<g^DaRITNU|
z7n>YY3WUg7T2`Rdy#nL~2oWHSr#E;cNn{-%kO(4?8ELVZr>&QeBC+fY6ap#NE6vq1
zF$?Tm$hF5nOD2Jms+eTlS|I|G?r};tYNH#Fs?kEkov)WFa#+Q61W&OMsRY6dKm(zJ
z4b;RGA^IHfZM@^pim;8)#(;RFiL01c-qwv2bGuIo|NB$~NUHrL&FH4s1hR(f+ZZTU
z%Vdf+bs<+fitff)@41l=R$u#4UDxsFR?dV}uC*yjVba7^S%|ljw`7>DRdlUMB}-pw
zBdAOc9+m-gdPjh)(_?VaEE8ti>C`3@vPf5yU6|zc5{qOHSLa9Ka3jlbo|Ch0Jp1b3
zz5e>^N6*dv-by?4zED=hv|(Pd@@a9E;oO>;B(xA*v?+^HU0BZqo%X!~@*=P?eBJ@M
zjXC5|FHB*>Jf}L-j&qtIvs@SbjCsm?PrbC8#>bxcWdw2xL{b&7&g-0lnlEW$XhO_e
z^X2JlQv_r@aQ7Yu{#zVm08**=>A5jUn%|yI`V*D3SZgw~IO`<Z*|+mB^wsw1)A70f
z!Iw_}v8BQ$oUz_(yC(6f9(VmBq@~)y)n!Q*m;g!^pqTCia87pK0_5N!>8KB>D5m*Z
zB%7Q=Rc)4-Pp|e~_u}@<OCaTRI-Lw$|G*UNM158Elao@*jdtZ3eYotFVux@h6avDk
z1d^!$oXBL{hK>v99bY(-)y6%#*~XMZ-DjYQZD5l`!fK|A{g+@2`J{w*o?3%ecu3JZ
z5_3V@))EZ{SqKKs@^q84a}WrCftT30MKiDorje8#vovcngq)~T%%RTC46`@B4tHF;
z62AXs&@EJ%Zb6k2F;!5-K!hZ)0U50faWs4GI{|)CD&4733a~22U#wkf7IVlEVrE9j
zO~kiO^_7>o$F(fExtBomww*W2mPF1}TQr0U5_N_^b~*J!6c8R6$Me{Y{u-8PAd9D2
zf|yK*?y(Iy2guu$2`MZgw72h%t1oh?!U!6R(B-r&=F;{Wp-Bo&3S<c4iA>IJ;DUtF
zalGwfVgNI!a|(nwUuGIF8*(cNT0Es~oDnKKS22f6Jaw-CzvfmPJr2({&t7bLEt8Pz
zV7aPi&@RX+RVfhKwSpHN{V2c|S08<`kaLJc5*{)OYFe5YiNT4Mw=oiy$=djJ<np}&
z+`h>7(|Bk^3alm1U`4<iBP&G7EiqI;j(Ig|zr#NCfH;vXLaRy+4vDzkgepR+`Vt~}
zts$of+t|Lw18~Eyx%qSpTG1}MvfR>dsX~fbmN}`#;1Yak3Q#9sg5fgSf!B1zKoCJ@
z2ZFDcpy~(PD%b=oCbF70WidZKIVz8O_wLEGfABz5|L5slV=}Bu?Td%K&}}ZK3G0gq
z^Q)m(Nky3s5fHSuctt^qDSMB*j-Uc$v4bGsC1lCLJC4qJe<=N=q)a6z8_stgIXrvi
znWG27>}$i#gF{jz9j2n~<#XjUYpHrQ97rD(R8W$l1L)>jCBWol5*XKDJi<**w`L#^
z!hDeeFe4>AyDn*nDZF`!%O8STxb-AdK(k0yv{q<kL#j38S<1ExvlvZ49a{h{6S!>!
z0<0n_B6(*l#*-(^ndKBV5tq~_4>5+I*gn98`_<5O^lbtmG&z!EQYsRtjw%yQji3oJ
zjlu*%X*iD4EXGiPW^`P<&RGG^O6+3YE!qsLS^fr(0NterCgCRlsf#IEtRWBs!O0<P
z#xLckF`;@IWpM(1H=Y1B!6CJD)FB0`<PgX~B($F93HKaRDocF)@_yeG{DM9a2$lp3
zJAfveFIp_tr=exqP4HraCDaMLrlVa1TKf1TWlp)G2~|tTgd_{YD`HcJ7(&>r_Sl7`
zm+n^r0IK<7NtAI76UA0Y+BZ4kG%ZyIQgY^5BQ|62@n(9u3A%+$IRQ~ms@_IYuV^aY
za@#8ED_xV}z2?L~21Y#abR!P_uV43*gW;&3Nwc+lT%zmp`qkpZPL_iA6P3xzmY`Oy
zO(wv>JTS&$<ee){BGWEcZ2%`4RD?QYkWjCQLcsO_o`?7DodRVxN2ORs2t<XZ$s4Vy
zGy<ZCS_~vS606P5Eq6d&b{ZPVk&ux<idkY(Qdq@AO(J{r&Eift`zqYJSAYx7tV|v(
z2VzC233Wa1n{B(nuo3JCk;pn{BwBK?fFk4e*OY)ZLx@BoVLXRfZmQ_S6nxE=D&DpK
z_fG-XD(*a+HYLw-^zvdn-GykA!J~<L(kMVc@f|FwIYHzuev%Ujj6kNzk#^zU-W~3L
zz^(f>0`s}GDF-0NY8{#cO&=kRNNOGt9bqmo;;|O|jwiO#I+Dh)WyC#GO$5zd2<ti|
zNVSuQJTeB97k7LA9zC!a!ySk}Mq=%`X|wNeffT~d9g<yOwsXNp>%?eFfe>RN)rn-F
z%Vt$sio~VM4AHwsAh+)qAT3URpw5<ZUgbbUbtWuUfrw@lV=X(!D7XTeNwb(52xGjN
za)Q`oqT;`#mY~sclRR$c-kt9GC+_V2J9|-fRy@+AS!|PUt8JUNy*0?v01J>bSGy>b
z{m~Z+$pbNs?jI0|IVT;x-q_R?O+5=Wqzcy9BLMrC*e%AL@e5KMJ--}inQdb>&D-b#
zq(u&yklSb!umB_N=TYvsOU8s<WZ4j6Y|CAkNuVBAD<}3&QP0OOA@|t<+_v#_b)zF$
zJjFoFc)%jU)`%7Aa#ssLcdv=Gr>{zN9TVEcl$EB|bUI_~J$rHY*}Y3V=odZu+>@(x
z=kL}QTtb^Pw}d?YuEq>03A=nk3}IBkly@pPo{|wf2AkU=Y^d|)iY6LL-hb*peQ)>S
zZ-895=U6uQ>34tq%pIu1^QM{S7$M}|cVna+Q-}*>{1R@#c3Yw)VS%)>N8p*6ZN$lh
z7zijf+yBN(_%nJN*n1g1dzr5S>|NOV>fUAfweilY&+MkkLqu1$NdzeS1TiGdv|}b>
zCuOl$4_R!2nwn0ss2)`%n<<FP!L)=RkOc)c`|te!K)zV4|NMxrB;;Qm@s<r~xc)we
z*ESDv9j^c5w|=2I=pgv~z)o&Nbt(E0meZjK7D;ii9ZDww0yChZ!g^y>1@at7A-5SM
z-?3Er(VGd_>vbppvzOT$Cw%nk@&(}Le|<Wcc0i)7CBiy~S(r6sa&B{kczFk0qoaEs
z)Kf4)OQlR`0EiZHAR%cN!sa*s=rZu=cWw*6+WYFSklSCqaGI=wC(sKIMc~%=pH2o4
zb53!Gw&qYbty_p;m3Q6{3g*GEo}@8vBqu<lr%@$m9w6q`>EZLn*DnE&efdZ4y-yE(
z^Vwgqcl%GvvMkGs7j73k!#{rT0&w%yClMfo7*i%|U8MlEOTbwMHsh8O0onV~FqxSC
zy{@rsBnu*w-NsxkvmhlSoX|Y;GVs_FmzhtS@aG(zzkMFghyUf=`SV{I2><uVmxhxM
z-uU(vxc<Q-f)1OhFl%RF$e~xDOot;7a7{I@_0+<sp+us1l|0}Z9OyD0bFxihr-#qJ
z1pL{pUlQ<(w;tgFcU&_JeqsBU*u>j^`vUOJ7ZS!CG6bh3CTLRC5DZx698TwTueJ&J
zs$4n!Qb48%m<B|AKA9GymCA4a;{cDni{s@Pz4GqS;e96P_CLGw*{^gOI@-Uu0>An8
zsFu<n+JR_Ng`)2lsW>^(luu4ZXRav;lw>3z8VXJi;=D7C`P5+Y#@7uV`#dP5y$`Sp
z55y&x_y()(ikhFi0RH{iXloP=24oU3hGn~1lC><yASJE|(jib$sc*9-LL{ZpJmwTq
zLjp8E^)fv4M5`F6`Ndlg=ECiv_cmh7y5Wz1@;u!9b@H9HNj4_iB?DFMsSdyvNMS)3
zcX(@I?T9UQ5fM2jsB_-=EpNUA@|TyhL0|(V|K<MiZ=nPhIWKej$N!hYJ5TSdEisbY
zlp!UWK#Jtk2+i=|q6vwbq*jTNfT#v@Ahi)HjP=)n|M+7Kbp+yXd0%hd7arceB+odD
zb(^Hl_W3^pzJFV10C}^bjX8CzSS5i-1IepQigbOTu!y8gOu8Xe6q)VJww^a3vxRoI
z0FV8s(!3e5XY+6J!;5#vm)Cgo61nx+JQ1z^*$Z&<PcHWWVNUi`2dNX$o2JQeOY{(c
zqSU?G0Q+29oP2J5u2~`n?Pl4PlN29Li{cs+4&;sB4m|XQc{kY}FVui?dUG@U!-e5=
z{F_(g-_Ds1(_*U5yx6w?_6ooGb}u?)UoS$@&x&gsF+blgOAiy9NR}pwz~0;?%k@b)
zlw#WFYXxP|`;>B3l%AFoCX%1LjQk(J)Pv-Dv~{4LKJ>*a`TyNXAU7}ZtrKjR4A*8W
z4XG%K|8X6DXT^Yc>Eix&$%**%D!s0anIMz`5LDQvEvqsS(QQfS(k!G+6GFonC%yR+
zJbEcMphhM_UhTdAJh#3lrAnzky~INWT|?S9F|G1?-QRiv-hE1eJ_nLUHZfC%WlS+@
zn@=5_ybsy%uviHxYFDgkbX&}}wWkmQq?o_{GW$<hmQ*@pTu;V7Q1ibL#SQkqV`Ob2
ztvFFq%;l?U{%Vd;rkI16(IAOb?D#LI1c%TAxj>|X?9xRmEz7p^eoWOw-h3Ip^Gi}J
zJLNRKehvSYf_I(4{tL!!k#fy!6{1@v@cFO7yHAgYB30@<1|)mzlRyKhhP+7x)oqku
z;;l#eSqOQvBx~$!o&!0NzYY)mC=$H8H)t}tZ@sWX>!0s%>&jwbMo!6MB8AtU$6rk)
zunj?DOi?OI+u7D!B@*+v-K%XlKAgTK=Q^c5V`taL=T{F;?dpv@JzOusOy2l8e5X|}
z($a{^`H|h!Jv<q1e)aLI)Bo`J!f;DNk@P6ITA)WY#rNOh_}kYY9j4--?HAHnoW3QM
zK<PmWQDMX0a>CTt%j1)?Yd?EU`h`kyR1W86<)k|lBtLnDhmIn$ri;DV`LuI~$|(#V
z{AM^m48RpVzY>dmc@%v*L~_Y*|KbY2>(u`FV12jnw`X&ix_0v{xHKv@Y08s6>>Tob
zW-_<sGKK`pMro0`VE*PS@Rt!%mSwvLfod1dB^#ZZ%RqNId&ZaBY_Z=O_!i<n!MjhQ
zmSd*fE>hk!o@xTDsMjd0GzKx6HbdA}&D>X5WGgkq@hI~rjE8;<<d9dS*|;%BzLo7F
zL-RCSvaRQ?p<2O$ED9k%kIw<gVP3`cJj9qZlb|L=&{|-x9w<sgDMHCQ%2CN3!qsHJ
z4-(&5D-=aIr{%R=SlsRwp_?A-75=x^bg;8B^!k={#oKb|Z_}HC<oDm=ACFZKJ?d2_
z&@PW3-UTQK13C&I2u)v(Gg<;PU1ur`@W$&r^w!cQcslE;ueOEFH>U#j#cKoYRN0sf
zdR26|vFPov=%dX9ygf7IzvuyGU1xqkX}#`1+1-D9Z!Tj3sRS210D1MbFV)M<suaEc
z<txmkn7Fu1+U3SXt&9C>?DPHx^0#|R3-M~*9d2V<Xfjdt^-R3@(<}VW6q@(G`lb5Z
zYg;{vr%`m;VRQ1VK@Q}QL#l!)G+%q<l@9q@{6UE8TPDw6s&lGyFe9TD@(u(Hnf-Tw
z7up0_a1e^^tX#A3wNxR8_8s;v8tLEMdc`Q8^B>kJ0;}aN@MaVs#XRP&1bm&J`<pS?
zkiUM{+TZceEM`Jeb_kLQndOLh2J$GtkJSn~apto%i?ve&DKCTP#ml^~{lzoi+YR<t
zKaA}uJ7?TWx>pY<-P9(Ciw-jT!cR`4%hx~H)Nk<}VKr~&U3tFHZW+!D^Mf|3fkzA1
zfiKErc67djcy(Tcle0ZtZ_3$_bp7UazWDz<^`B3b<~RTT7fXP&*=6oH-m3?g1twV#
zKVkFk8wOHNCj(sn55G8b^0v~I#S!N>H?A&M#{<W`9+?ht+b`^2B_2AOiC7m6tDEih
zv-H7>v+Kk4MzrJL<naIV5+LPtVCO9hT8?>Jo5<cAm<}UhMBwdjTt><X$aFIB`nMmu
zkV?vC>(P&n>ori@@@idPE#ID<Tlp8S@Xf1p(y|D0cAiy^ZR^GPYu%DfZR*E=R{i?#
zpqx%%I-Lys_Me^T0L+TDk|q=OzBJfmax?*w%_M*2<vSphOb34Z?)t*eltp2tM(0E;
z=J=FyN=eyqgat(jG7&tSnU}NvME#sJOW72|F!}q>AKgjt$z;f1`^6MM*3r(C6WIHc
zVZty$2QQiO#&3taAQ;mj|L^VL<Z8JtyB)Z;U5aP`o#-HXb~xgXmgV$Bis^K6A_$1~
z-C>*0ueE4TmU;Q#?nyApWcba$IsxoV-<^{zN%u<10T@VPnxDLUZ$6UA@Y{pQP_pR&
zf-Fm=oXBJ{MPxdWaFF<;2$Nw7GC9$cazMJ!`A{lenr;7Skb9`CGVs1J;S6O-DlrY^
zDWtO0$eXX+H$HpZ9DRy0l#8X7<%ku?9boFhX(kj@l(Lj%j?5FRGdbG*IFe!)beAHo
zAbL2BUmS-gOzvlRD&J3#+NuR&`M^Ja^ixnL=>Sie&vXArL~qSxhmfA8ipgY>=M}kl
z5cU#oJq<Cp84zy2_Q~V_?sCbpkeqW#*nj!-jolrZ&iB(@j}X}(0b8%uMa`QZe!tol
zi0de;KhgY!FFZ&Bh6ky_<)2(G+=!`m@{_L!cPDuK#<!l_x^7Yg7xMl{nCLfQo~kEy
zzixWZ6#RG=Nu)5x_x{E{{Eff+A-4E;)<KoJ0q#txoxau8iXLkt-uHhTB={V`H*<V;
z`s<w=c(-G=-?~5+s8riOvQMD`-1<KkDqG6|_Fw+6(g@$TJgUP2??d0=IJ3pNeB;s`
z$m0%%j`|E1kydX%@iL$4x%uPE<*aY3cn6igUjQsnjpx}~gF|K=Y*mSTw34qAZ36lA
zD}2i5-HRFu+Qc^~?gGiX#c;j#G`Wo_uOa4rH3EfM47qw|_r*&<%5Fq+`PD2qEbSXF
z-IFT#Jx_;=&=>tZlX=&2fEpemF)v8fHUO(kq+RjH4{pM^+ovuB9`O%h(;xkm@Lv*3
zdoZzhcV)M$1Oz+=4Um`8ug63|{P3@5ZbhnyzJCe%bSt3Tx==!7|L!O3UM2YDZXn4!
zYqVxjvkLG4@jQ*Mt7M%b*r%HS5Zuu*8^5d9=p665ZkH{xk<-H|2R_XfsUfT{ee^eU
z@E|NsYd^p{yU(_Llnhz#SUbP>RXY>EN$${R3m^+)i9R^N_fVc!8sRo$mchVq=<tu1
z;BN_fI-c0Q>kkfLwXIrda<TROky+@f7Vh7Lr}D?k#8oKoSxWyzC9rY*s&|X=MI|NP
z7oE8p==%{o6583e8iDbxg>?F*ySqp2X_b#e<$-DZWSba?#qK@(&~@e}BYTV6O0$Ak
z>F69+z23<R+&$cpJHzB;${+X$JC+7^H?<Z|=^~7Tf;MhLFN898MtyhIAAH(-YlP%p
zAwT}{u*0LpcHp~$?LO_KI@0u{BB-Nob&L%U?Ae{+QONne$&m+)>nA_6&S2G@zBvdH
zx49I-gM+bkYbB1El4p0T9E{JN=A`R;zP^G|-e+7tatRcEwVQEE<2!c<x&_N1mS<EF
z1YsSM=F-O{_{Gl66L|8@_r5b$S?(Bwr)<KHf9%euWM^bW$3LWyGfvVVp;KqES!|W!
z!bk3}%b>~sP#x4!%F&n<iX?$-h|A~-2Be=m1TegmS+7~Sz(<r~$r~2=pLl%yBcLUM
z>o^OZM>1vhMQ$NrO1q|wDcrAxep>Lg|47rMKYaY(H4U1zL}B~v0v`J}<7O@>l%fgC
zu#JhJWTJ}qNY5|t9>aG{@vUE9eD0eD>|a1cDMv=y=r!C?IJuwDWB%x~``^H!0L?!)
zxb+$rD3pSy!5Irv*dIxiDX!ty^p6nVfFA>UPrQN>;_Q{;ombzHzcS#_=RT!R_A-Qj
zSbygw0FRe{JRW4_-pecu5VbbsP0$nx2q_m)W?0nQ#B}=tzk84P`B#txxd7h5|BHPp
z`+G-k{WHm5M%Y5n?QVSD0SK2vZ!4(hzKZ045<?)HQ(W|mq26s8g}qeAH!kfCv&bJ{
zcwpO&dbCCMU?+gFpMX7M2~AQl+R(26%k5Hp<fFwmbyNU(s;1cy=eW84!&LB(M{g|N
zEXaR)_phH;zRaQ^MM9P0#&ReWFqy5YWFKIOQ2p6+PZ56ot-tyu#3#@ZlK!dvA>-d3
zQJKdsu!i^U?CyV-;1XXUW%Ia=9<1J{>>&C!_8HWkZecUh-8K>4Sn-U#_}yJ0)~9{q
zNQdCX%?G@v<o9=_z5a6u_R!i99EF(JrvqGc>xy4|`TMhbjZVJsgZ)%oQu{-V!07zF
z`<|#>xJ>@LyZ`dUKdO2atnLa#MkW=AnW4@<aqA-Y#G7yY4292s^L`8WX~p^MFLid2
z|9XFh^<{Yq`n=hc+px~@)ZA#buw)d^`~TS>zEk)7Ws_$L<JA|w_1E{FbM`5Yf)`(S
z>?x3c;^uvJ{C{Ep7oC+U8ktHUocbs(vg!DMn_o@Xh{WedU*pj=gj+xUS$gr#G~E18
z#U~07Uflfh){kRn#q+#!j|7WL{D%lBq__&X?a6!d5xMR9;?Vl(<jE_me>b{<L*9Qc
z$zRB(JJ5MO8Hy`={~Un-%s&i=zjyfS=`)3t2f9|7yz>@se&uw^{^W-|bfN=>>-?yX
zlcYf~d{fmvk{Lfx|6Z**(eUWa%RIYww(E9vS1<e~-2BR?SOX8_1dqJH&7bTnLHsH5
ze=N*bTE%tN#U@7yZt7vawV;q4e-8id<uCl3^11la%RGi!%07z^tonbs{K-BktYYO|
zc=LH~{^z}aQUA$T{YS=MuffaNR!F)oQ(|AiC8K6Q?#lj!m-yWd_1J&Lx=diZy;~$I
z0^I!ar`iT-zNslbQ1gHIhh2iL(=Ib>W72*T(_A-!k#3k5I!1&3{tKs{f9UyyhBD{-
z^!R;%|N6-r!922&)RF(}t|fdD_mJ?0vI>oEm<NM8CRmLb(GGv}#Sshq)^|+(=S`0^
z_Z~y~JaF@oPYw(m@Z6Z>$Nr-N%wMd;Hl9H01<d9}3Hxdzn~&Y@HZ8j6PJy*<WEtJ>
zZ+rmoU*m;Of(tikO!FfC-nRg}6AD{Ny@|Mo(bo+I+2dll`;sXEUDI_<1MUet+x_ui
z@%*RLg6$aI|JJ~pAMJv3%O0~fDTML+<^XJEBPr}ss<&V3poIAnH!`6Pet*;|4}A(P
zaI6@k3Aw$3`kOWe_c@e}B`T7RT#>OxeooA`s7GQkj$uFj-fhdPNX8eN?*1U;Q)q#p
zyEDU+b-(zV?~T0-ST`mT>n&p*zS#zFBilyiVKs7k)6YBa*5<lOBxpQ6*gj&Mf=}Fp
zr`TxDn9uNzltsPJt>@8LX<dX8DX@(#Q}VIXg^P_J)&MfsIF_*9JHO8ZH$O>$Wj}_e
zd5mj1%UQ`1%xz`au^A)@4kf*kG{%wD>%PgA>_R!;DB;1VdDMbWgzIuncgw0Eg=7SS
zHU@1Y8ge`Gh%7DVOSOFGYz}J!s0xL8ZJGR_#y$TD2f;XF@ea}jP^Frb3l}y~D@lnE
zWgIiI5+xBs8cjjD1!y~svGI=z%svHZYulX}AdE@wRvQl^R**wzs+j2V$ZjLYoQUnN
zV0G7l#hh(QQTcFOcJovdf6tSv)A6%LfQ3dXQj|Ooi8`TPZV6+njTJe|tm%v8ZfM|_
zEWMUiPm2H^n1JU$L4bKWE$L-PkR}u<gl6pS(bQw%PY}%+7a=5x(Z(9ASOG%nEaojA
z0=+-Y1dO%!;}j%C44P&etBlW$Xmd9Qnotm0wuQt>`ml<`9I4tKMLcS;V#bHSqfapb
zG9E^exC0CW)S?OIbm}Xe36h`=facA-u~5;J5)&y%E*(tVR{v8Purq<9^S9eUM$U^_
z*@jtJ&eqF~r6N_$NNm<nv+Prv#cmUTypBHVZ4?^;4gZ5CVEoQ8MaU9~!2YNa9XZ*?
zau%5p5hp0%W=R$xOHBl&3W<y0Rl8Ug#g0_^Q4`R)`Q5HXA99To5oz_bVgO8LgaWI@
z3Y8RmTkIRq&stv+dJAN-v6f=QalCcOA7ulcKTQoNAI_WI4_ZeXEA7YH5w&cQCN9KN
z)u6s#b~&(RZpI=s%M}EwRt;`dO^v5eC9-0l`S1jM_m1!O(SUKZRmgYIvP#TTOl&uG
zgDhg6YgdMjz)Y)7soJJqjd)1~g2%XFr8z<UK&~L(5#|0L?)QpwYh>Dh7zJQnSj{|a
z0<kP{lrHU$6yC5zw>~G1y>r@P?9^aeaYCl~llxx%(1lMl#Eua12#N{tRO>R|C=F1r
zXk$|(53{-pr&z3;`;0aPO+j4RH-Qke=yfC~-#L!<KiLS5S(D93##44aefF{sL}-fa
zx$cV+Wh4_jVm{LvugaVP9<u{KPAK&Ah$imFgD*aIA8h^6M!|uSn4F-SCjcaM(&c3m
zgmy_*Ei35vdsAf14BO87MT$XkS$J5y2_feOwGC|QdyVlan|1%i-+vlrT3YQ`6v!5q
zYt_Ds2_Yv`L*DA93~0`2vDl&&o<c~A*s<;;E0!gS5jXHZ0aQ|nX0o{?*fWqKl!Yb8
zz2c<Jde+P(DBFOGizHoAHKatlC_G;TJS|QK|ARpiY)o@XH5Mso<IQS2k|8vghPo)5
z07~5Yat&*iJE~mTwh?Y!h<6dwKNPNZwmB(vblV?yRIoZ!HI|VOZMIm?MH>wUo?X(y
zm3~DXV!&mV>2VEps`K&>hN=yaoM9W9gkcns#z{G?f{|t^RkkjRoI(oBY`qrC>-pR_
zI&U%szSv6m^_bg9I!2B12MBO^Yz;zP6|oL=hNmDdLIsOVkUmQ*(WTXjz_##32eVaD
z)eT6`y!DuYS5rsq%W)a@?=Nrs{m%uUrja6$o2LO9TUVg%j<uE-sr0x_5@79HpCOBx
z&4IYeVS}&C6{9CDJKcFC{^^&%&1n>0Tdt!cEsBSLM##%pR%~&h3d`WGx{2+w4;mYY
zqGWPl{>K;b7@?UfUIY5k$?%V$P>d)BNH4RNy-Vyj%wd&Bk{#=!6{0|hIf$?oSXZPJ
zQfxaN1H>l5tqP(RZCoatiywRNKk&T5{urKm7eiz{o@86-=Qdm7E^Sq)<DRepWa)_^
z&oyIiYek4ecJQ>VGR=<C^QorbmT}QxwHymPGw@{X_>z93<$ATC%IzF;Pm%dP=q&X!
zQrtvat^_ewK&o4KHAiCDBsEi0wRe8-)lbSAmp|wbbvp65mS)++Z@k3J1VMBoeJ1$2
zZc?;TmZZf#=Eh7@)sS?hTX)4svM^V(898_#RKfY?OLRZ@G;{DM28%Yw)m*@lEv6V2
ztcz<GNn>*kmzeg>A55&|S~)G|7=w5`I>$g{I=gVi|MeigUUp{vIi}&Q;nSs!|D6vE
z>&Q_&Sjqd%@KC3d6ep#{Ys>DuptF#e2#hRRk#g%Z$UTu9Qb=R66TxR7F)eCxhw-gX
zl`@gP20T@(A|#@Zc95LCDLkibgLg(O#8?}Jd==}8(p$$%-xggYiN_S%Nc#?tE=W~K
zGfiQsCA|BKXD%VG(%prw-AbqT()+JmX8+a~B6M!KjSZ#D(#4l};g<?cz`f3?Zo*;}
zq9Zm7bo<`3%xS^0>)a^21WRkjvRt)3UC2VFl?qIG^x2yoAL|C(>PG+6F^HLZ(>ac8
zl^o_XkWC+R$m=?d)o^)Tgi>|WdZ=<&t3(J<AkV3U&tK%$(-vh_&)t*b82<9_ITiMG
z^iK^XATe+8{g>IlTrz@kYg>kGjIr7EUs#I-NE+uBQr&Dx3DUw;XA4=4m6;Y0SU_WR
zZnt3n<6VP`@UmMs#@CXzvBsf@9c&V-<Ro-stGPvYXqM3fDT<-gEb|gVC27`J+e#$7
z3g5S}4p_%b$!A`M@94+4C(c7~^KX5ZI8P|(H{huf*v4%Li+Rq>En<kUh}qdE?ouEd
z%peD+4cV8OHo6*kZgwfd?)GP2f`{x5gz=Hl$UCzw+QBB%ge4$r0itzIn>93bd5X%4
zq-w7oXv!mPx*I1DJIa1(IrbfsLS`4P>^)>(_!o5@e*@h~_rG3&`R_?q-2UPf_y?OD
z44bP@e5tOB*EziM&=3c*OgeD6(FJ6h;+Ui9&W3$64Je^n%feb%`6iRp`qE9`l+V9^
z@drPJ^0{|^nX_C1o_g=&1=#z^F0HlsLh-{6KKtCeKPfM44cZ7xPi5A26Z5VMqn5Du
zCG60r-utjfH!{)6t={m5ho!8CP51Bq;pWc=V0it5r`GJ{0lxV5feph1Y#1gsp`So^
z{l@Vy+1&V`ID)+cn|=PLKOcbO*WZ8giD7_7G#fY0y>+h5Y^_AxruMnDvN!i@>Z^I?
zFwM^#ZdI?zNsd?Js`G`~8#hnY?fP&3ZQ9!aU+hkrP8f#VU7Ijm(P6mWU+p*j^~t(>
z|2pKqSbgD6p7r`aec|t?P?}0H^yc$TZC;*C;6PMnZhMa(m$>%D(YDYd&e||6Ecl5U
z^Mka1)5gwMlY#uZ{L~eYiLBt*r>q9ql)3B5#*T4(Fh|Y|@VoGa!LDCc{@n*9<dVj2
z<i&OqW){v(pq~Qblf6=aa8)^X)xxn(-d?nNax4bZYyQ-Z>5fbXu0Qf~U_R!>hpq#s
z@9x`@p;{&wzIA!WbCGil*S~Q^b7)j8GU;8q5LYD}jP|VG`_cgY^>YlzCn5HoiTU93
zL*4pe_k|zb<xqa+%y1*ai<`3scam5REbX48>O24bb+~fpV}neF)wes=!lX=-YXl^h
zg$Uw8?x)keIrwBaVmm!JE@uqI`A*`szH{dN=hIydr^}Fac<9X&D~J2-(0%{o&7UW|
zmmnXG`JKs`vq$swVzTu`w=DYUuF<kgVDCKi$BLXi@rRF(hH!Qjw<Z}vKKyU@p)b?n
zgzp67#>aZc!RwAM{{0|vxc6hn-ySAu>3(wLsP@yz9e!&+Sq12H%e+e+h3~LMuV9|7
zdf?nqf9#Si9>C7;=OH#4e)1jePcEzfWCg1>bj;&=9HAqV9A^=rBua`w<}0g+&}Jm+
zhW(cxAkhl%(uL|SMl*I*+X3T2nSo|`kjyL?Z5<Yy<<in3>Bdr1BCaCkDui8LOCmXh
zRIjbfuu3L;eK$|t9n-IV@lT=FF=F0LSKJZq@)oN}eEmaI6TNj2B9hh>DYV-n*0?~z
zqwjrbSgGVNwaIi=7A{yUEjaPLW++Ftz{#Xzm;y5V*3_BU-FP`}JD8vAvxf{F>@~x=
z2VKYEUGmA*xXoL~rD`>)O=Y5Xtz>VGPu61HlrzbX?Br1UgDsBjZT1d!IzAcTWCDBF
z<<GBlX7^~FIdI%T?8Z*Fdz*V301LoF*KZyi9>wjM8yQ$^&rH>)V3SFnt^$-nHL7kT
zS_liw$T`9{n2l62<-%?^+}7K(F~v&&%CT=tE#UFAZ}-=^mtD{U+|rv*OTos71j5EV
zi)~P6Zgqea3Pp-*VpX;62x^91->2KVz+89zpMJ6)dv;|Zc4G^PMj|9af++X1kL?b2
zd`f1lB-N5Y=VaM9?474fR>*pGW8mTbp!1<#S#aNLP7L-olngOV8P*fnzkc)J%rGv0
zT!W$KJ3ZDu+O7}5{#z&aE<T3?a5B7m{qSKK#1HL!SIW7yDY#7iWD=A#i@ig)C^khO
z31dc6Y>#_79^Gl&ku8jxpFB`Y-_Vo%E{J9O_xfnwk00*YW?fXa-L7I&SgulI!ELJq
zqY9|dmmMnQnw-ZTUAyO-BR6cE*xP#RA{i26Y_*a~@OV1W0Y1ZL?xWjm09<p~VJX!!
zWG#|ItF@<orsSYVkx~;~)o51COBI~_1djk0dBxw`z3amGrF2x1T&5YmNPY*y+mi36
z$B)jV0^~)JOhUG5k(Zemp;QeR!H`b#(iP^SzEJui5O%tJ2`=2?0)}yEgkE;6X&eZv
z2dUV1ib_t|Mq<E4*SS^#nuI3Iz0*u#xb6DTN!QOu()hM1+{VUi7sj_-KD9c20WOc#
zuqeTHT`fQUc}!1{{L5f-aur%hA(BSbP(=FxmN(_xHY?3*p&R+o;S_5s%e$w}Wf0nR
z5y8kG1>S{tFO~gP34eKtX=F<wrCcm#ZGmVugq&wZP?V*uV$eYJuxJ^r{kQ^m+6Eze
zV+*den{j{dtb5(!)8Y5cFY^9rSK=OQ6LTbjRjarz!aOH}unhn!jTvDLY;kDDrD1Q^
zYUDED`Ce*soj#sSxdW=W#BIIXF{%v2lmn}pxo2NNMMuy8gt!qaHr*nJxTzL(co6Tn
zz&D>O_wJ#Rg=WrO7+3BrQobWs0q?8dwIkMNYc*`Pb>6O|rnHt$5)vR6smvm3L2?8A
zrh<IB?Z4W+2XOgkd#9ukFSntJn=vT|;Z8#*jML^w`$dAA@9SNMLT6%&Ib~!$Tc!r>
zs$j(u3)#bZS%$KkFENO%&l7~rP6>a&tuLcouy<b>8-qHl7~4Q%q@8H6*<JT!$HTjs
z0N%NGJ`hp^o4jnjs#6o!n<&W%I19_9a+hrMc_VlDtG~WC0V0>-0$+g@+DITe4jT_)
z*R2M~xZVxm{T}4jsEgWV-V)Q6P$L1n*MK3Vcy7k#n6XHD*Xg|dO~!`Wmv@O0XSVmO
zH>QiIX?faLDuI_F!dJ(^PTREHUoz5!Bs39XXNp8!cd7&xa!M>~Ogx6PBfso&rLW0F
zZUGl~^z>l;^7o#*kbCl6bBGaqQ4v(joc5mk!EWS-UI2LP-sDykc8yW;T2#mRDU$@s
zj#&1xLPgp%HY*Oud;AKx;JdVke3{Fmp1tPZ{q=K;l)Qu$z&gdkv<c7s@ptKaE&;b*
zeICMRF70<0clM<3ovcnx82k4cGSeof5GaG|Y_;4#s<PL+vH{t75WgC2!L!Z<*t;xm
z0PlSF&FA83wa&Z7n{Aihc<$-%z6S68dYt!Le+z1SgI7nZwEv5u3}L+-LEi*XoAEM*
z{hjL;QHTw`v3bb3-jS#FdABUzRW7>;dlxR-Lw?~B@XmjF^f~`+8cC4wHGTKHr%UwS
zXLi}eXD;y)@7$?5E(*kDi4|1jke4Z}t1Wx~yw7VtnMjfeSq3KOj*p3{)T;T{N4$Mz
znELR~uS_qbN8kQ(IBfO~Z(JV$uD|}9ljrL3x&1$WzN4-E>meUc0K%D*VR+^W+&%*{
zo9lGVwJH_sV!B!isT5IFzZ&*~S|sTvnx{okwDYvExHGtX_4coL{~NEsvp<k8@9}qE
z;c0&UXaBgn2VwJIwdZ3`{q2)aynOKzaO>j#*Z5mC`3&%t)l+R#Z?eR#jQt;$Y_>WJ
zA-)+fNL5x$5K>%KtC-dm2tV0l4{pAO@k8S>T)524>Yu&$;zvTZ^5PS(WaQqo#y9zE
zf&EJYy6WS)wJKR+0cf-U#2^S6LPO)8V)AuJcyiu{3vl6i458z*d+=-9e^04z|KPE|
z`j80nt-t!gW!U@mNV(+y`b*%|(InXZkFl~KS5+wS1S$79IvEWlC(5=95!Tv*Kt#Zw
zzxBw-;N1U}A{Q=j`w7G>uZZvd=+9ofx5!35`{G~y?1u_i)SDLp`OGEYc6o_XzfKV|
zgf@cvMs!U?1W5+_vUCEPC4*R<s9I7?CMtZtxB$;_drw>qeszn#+!|7>k)N^mP58$I
zgnt9?{s2NB+sy5YFTXmTza643@ct#9+ID2!gsvmb8{6#{B3btyALGvT=R)sJ=b>{j
zPRhy6CV%_Jj*1lq;PCZ_&T&2H0R!l9(==92|7ddDG|drpCyV4!9{z{T@n5V(h3RBU
zKf(Xy<%g~vo)kwi6w`SZCZ)wxuvW6*^O44r$fK1eWT+7;$Dil>m+zLXaQPMrA?IXw
zC~{r*yDrqVpqV_j7F;~<ue$HDR*Jv>3g3xF8_Uo_oLPz~_92qTj!~#uaHo$sL`sM>
zwU82gz|D*IFqogZB+p=*V9dwM4+4=n%r-7#d!MLeBk=A;0eW$>&c0#$T|rZjK~}ca
zNCln=<*o?cV7aPY$d<YcluZRQ!=LAScO<wiFY%~>ZnkV3&^oea1JbV)T{be87FN9X
zynp{08&}zR?j@FMa_CaZx<a}HYYN})4E|VOkfmWoZNPjzFC>9Fe1N}vaaWwM|5Ns#
z!F|_)Qp4OgVSz&LVS}w=VQB%FKg7*fpXtVE@wva@p|3=P7;5Pjr7f$bTPE2mS&TJ%
zk9!a6$h=%@H$M(%%eO3uW8HMhSFYsW{N(_Met+Pv(m`{iZ_f@51}`}B(uQ*Ikb@8U
z_OWB&^U3g)E7$-2FBiWV<hkm51OL(d^=*|XIPWLt5Gm30IYH7PecXGT4t!8wFX#_$
z%&v7hEPMd^>ExY%d4<=z^YF8u!(+w7t}Y+WW)K-{TioE#1`u}`#L)qwoJ@ZDFB7jH
zJy%yhybS;BNSXAix;_-2LO#BxsGvisgB;3>G~mA7)QKvEDh2go1VQQ&xp~L1KK6`x
zw+fYQGZ5QMJf%+W`ktS%-~Tl4-L1)a?8|{Jp~Y&Mmf*tbE>k{#;ncAX8ajnKu9Z!<
zR-(eH!ykI#?)cbhY)2FURMM`hH0-R(9Y-+SK5bVe-+C(WyA89MM_G(snq**e>UQou
zj-Log=@0#38U~LRx@c<9Pbb{`>RB7gNB(^=z|i60;IKDP@5Q%yIhKz3`3{SwhXyd6
z$n{%ay*fE5ANhCh-*oG&FeR0{yD5qw@GuM@C3}x^rgt<S^s|#AnVvWnGV_D6gYR_0
zu=n`m^}@g{Irv~ey&72dd93aBsrdlLzyQ!sroxr4|H+r?NB;YHxPBCe6DEBrc~!{D
zAHEZ;Ox8>G9v=gW?(J%QB8S({THQHyVqzeRUM4UM!vNi|IUbj5S#}*zbitXQI1CKs
zFhVlaMPFQ<o=kMO{+p>B(p|@FSiapaq}eXJYw=;SiyaV{q!0<GAQV<sF>ga?ScW{a
zQ(E{Pis%T=1Uxt6MI&3{yxyRpCQ<|_#~L;_QU`<@@N6kY)-kpz=4Q?sLIM#Nstd69
zIMRV{52uqP*YpSv-6C9#DToVlIGjwz@jb<kP#MwYGX^+bc3l_ClVKB6c@!-UH#!Il
z13#JUJ~sFf+pavZ&2k10A1!1!BPA2O98z(9ca6wEzwv4AhtkMMdC(tk7lj{M$X(Y>
zhhccV>x``<9T=S*8o~7l!doX9TVB<PrlA;u`?eUsduKR4BC`hPu;?f~4ad<BrYr5H
zB%Kc0F7I4{r*$AL7PiQcZ9S{=I_<9Xh}O^mQS)*vI*yPOCq$7=$YZa|lt%j6F0Rd8
zle*lbT^osZS?8e2ijbC~i5P5L`~Zkmo?uQNjQ|TVt0h><%!8)Q=**kN#0<tdd6h-T
z$(K?QLYv3#F59lU2cphqkR*VlYL#2Jt%j7N<ZT4swYD7v*b3HZ>!}l@PL+|Fg%INA
zF0I{Y$9qVsF?3p2Nn-X?=L!}U$IA&)ke2yK-P$4Skmtluxnzqm7^me5yC4jB(utK3
zLykU^<f%63bDX7k5AGK83_`5eK$T{)Ml8k9h-H!|rHVqtD`vYplWK_FA|f6nv0l{m
zCS)aLJeiMfP5_vf28wH-)SO{O>x+;$y~YOx+GsK6HZ6g)v|NxSw3T22LP<&sGXz>0
z1EvPSjkxHLSsSfko)RoV?xXE$*T*{SRM0Axv-K?@M5(F;WDPmBcLAF=5XiFz0;{M2
z`poW<%gkD0PQkLYAXI=V#Tt_qNp8{mIVsnekw}{I#obu|JTgcjTq}zsm)PfR?1o93
zmtkf|mS_xh2ku;rc4%WOit}&j;PXL_mvY#9?T1Aw56eQkWj`BY0&YwXCtGntx4GPI
zO5u!#BRwwq!-H~SWBu6p{pM=~3M9nBs+ds&T{UYcWC75yTu{$L3S%ens0|k4rpU4?
z){BtM5-O{93tox^kUU$b%u*M|a-Em!BDXe+CZXtB*l}G)0m68_2)%})Y-LG`QW)wQ
zx*~0tBRo#i<nfL!A!#fjQHT@RW|keK&Y97aDO&{|k_S|zFWnbG7MPC8CiA=3ckNdA
z={(+O34xd#VQs`dY%}G2CM0D=4FnM(LTpn>CZ>|uR2o_%mJ1`d;4I?^o-K2*B)(cC
z*Hz!o)wUD~Bj@-g_X@DzDw3K)$E;3Tgy@<am(Xq45REVz0a*&|0&-|Vz~W|U3Ld+V
ziWG?%dhb&MIn&gIvY9J&!X0gNL8?%e>wNJ(0mg#(nkLcF+ibgOmLb&Hmw6kLV#F9=
zPRh-U9F{F!=R&qEtI%&*Byvn3Da57MhGHQugLPUhZQlBls1We93@ccdZ=Qm?yUX?D
zGSitzX~op9EsoAkbbXRJUvxM0sy^J^5Wl+UI9V8~Mq!ilel;8uySP2nX*1stx=<WN
zTO?JR9%wkH>%xv}oBC1FE7yuJ$z|#z6#bd<O#Q(1g7P#L-11T+EkbAw&(dwyqO)3!
zsuGCgG?)3By0yNsxTsmhHg3AKY{r|g#v1@kHWoIlBGjEV+nA!PV@R~*NTStXZ|DOv
zK={p}ciI->^VH7{uW_hb&Uc4b=lHZ-cZcERz$8xkP4Ld^O*y$S&BnIk?Sr$(pc9Kp
ztmJ&(A96n7C)yv%bT&Wo^PsCzF)i&($xJwNhZcC(&OME6C9XTX$SQ8O&SJ~HCI^~G
zyP0kAu*zsnXi%dGh$hcUTSQiYh)bW1l@cMY#(ump<P6IV#GEx)3h*W$7GNJDIh!Pr
zJ`uwr%r<KpXd0TOOM>|-2(u>Uz)F=>Ei+AV-R9O=R;kUBVyb-MmKS16G9)7>8zII-
zaFm>zi+3{&kKao|@^)E5U-S|=u;Ebk<lum<M@6?jaLghcGZ?^7a8eZMP}wdxE5qBx
zw%U61ZCMhAYu&?L*&k1&JdkUCuA6iZPc|o3O40%z)sGTDd|2ybGdVdtutivHPrRRZ
z(s!jykLDh1!4S9QTMVHXKy*5sUw1LAJNm)wq&)KNf`{Q`BKe@OgTpkObUt0lWICP7
z(Q$cDb|s2PdrwP*d+)lP{#~4M(EFs70_|cr?vID^q&V3`k?hl>*kTKY<uzx1JbC+E
zsyMmo(Ux5~xp7SFeQS2mS=palWzgRBaJX7ZcW@wS!6Z-uF3q!iv;ZuBUWg1F>DjhO
z=v6&9=cRQgi?tuLC%s#7w%Z<ujl($|&M`~pC%WzQcsP8!ILpI2-LSkSI-MNfSavo)
zu<apI77t6a2vgZz(%pRR-n+N&OV7x&ZAV#B3dBH|ht%VBY!`E8RgTQtm??;8ljhk2
zS(3-CE!(Qw&S?m7Lo8D&il)|rdLi4c327tiI|YdO3->{M_veS?VwW7vDTZ}rq+ALM
z>mWlfNH^Qok+_aql_ZivSS@8;)KaCmv8Ha~%Er^gv<WL?m8%d_mc_CkQ=8U2dVjv}
zfdo}}){*(P)D(lxC}-NBeHFIBY_Y}WxoJy;Yzu_ANK2v6oFuf&mQ9`O5N)F?6_$BU
zLu=j=)T|f>+kEW-i1WdF&t<t;>tr%8wd1ny%QM?zJ$tyB9?biGh+Q|C9CzkDf$14>
zFPo0@9%jK<u55lLFx%?T)%d*YPnIet2a}VE<M~k`U<d23J_xaX=;C#m7q6G}+7%Dm
zbYQBY2cf)vR9Y-jSROp=hZYSVrV|nOUeEQ@q}}wamw{_aCkMm%%+>N_I#@X^hWU&^
zu0|2^)zAKm58eI#5jVe2m-#p6m`ZbFh!gPsK&&in(6vOmgGs7U6V6F0!dgzsGvdc~
zHf{T}2VIXaJs{ZmJe>iO0B!%MDKe2H!t@~I_57<JQI+-)0w539YMZNgIUEl3+D{I|
zhZ9wW;bgesS515=o0D>L?B&%t2S@cv5II^3uH7|G7^d`RN}ISKbi-iwg)f#Ljj?}x
znTl;hcbwKtU*0KJPj~g$yZIQar$|?h|GcpfO;0adO$y9Eo|66l2gJ#{lrDPB0RR91
M07*qoM6N<$g2xhYasU7T
literal 0
HcmV?d00001
diff --git a/plugins/_orchestrator/webui/AGENTS.md b/plugins/_orchestrator/webui/AGENTS.md
new file mode 100644
index 000000000..511b836fe
--- /dev/null
+++ b/plugins/_orchestrator/webui/AGENTS.md
@@ -0,0 +1,44 @@
+# Orchestrator WebUI - AGENTS.md
+
+## Purpose
+
+- Provide the Settings > External Services configuration screen for Orchestrator.
+- Show command defaults, detailed auth state, Codex device login, and safe disconnect actions.
+
+## Ownership
+
+- Owns `config.html` and `orchestrator-store.js`.
+- Does not own core settings components, global OAuth UI, or terminal execution.
+
+## Local Contracts
+
+- Use the Agent Zero plugin settings modal pattern: local `config`, `$store.pluginSettingsPrototype`, and a gated Alpine store.
+- Register the Alpine store with `createStore("orchestratorStore", ...)`.
+- Use `callJsonApi()` for plugin API calls and A0 notification toasts for errors/info/success.
+- Refresh is a compact icon action consistent with Agent Zero settings UI.
+- Summary cards do not render connected/needs-login/not-installed pills; detailed auth state belongs inside the expanded card.
+- Only show `Connect` for adapters with `supports_device_login`; currently that is Codex.
+- Only show `Disconnect` when the API says `can_disconnect`.
+- Never render secret values. Auth text may show source type or credential path only when it is safe and already adapter-provided.
+
+## Work Guidance
+
+- Keep each agent row readable before expansion: title, id, description, action buttons, expand icon.
+- Put detailed fields inside the expanded panel and prefer a simple two-column grid that collapses to one column on mobile.
+- If a new config field is added in the UI, add a default in `default_config.yaml` and align README/skill guidance when relevant.
+- Browser-test visual changes in the live settings modal when possible.
+
+## Verification
+
+- JavaScript syntax:
+ ```bash
+ node --check plugins/_orchestrator/webui/orchestrator-store.js
+ ```
+- Status API smoke after WebUI/API coupling changes:
+ ```bash
+ docker exec 8dc967046cda bash -lc 'cd /a0 && /opt/venv-a0/bin/python plugins/_orchestrator/tests/test_status_adapters.py'
+ ```
+
+## Child DOX Index
+
+This folder has no child DOX files.
diff --git a/plugins/_orchestrator/webui/config.html b/plugins/_orchestrator/webui/config.html
new file mode 100644
index 000000000..3e4225151
--- /dev/null
+++ b/plugins/_orchestrator/webui/config.html
@@ -0,0 +1,504 @@
+<html>
+<head>
+ <title>Orchestrator
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+ expand_more
+
+
+
+
+
+
+
+
+
+
+ 2. Enter code:
+
+
+ Waiting for confirmation…
+
+
+
+
+
+
+
+ No agents registered.
+
+
+
+
+
+