Align the scheduler modal title and sidebar dropdown label with the welcome screen Scheduled Tasks card. Update the scheduler documentation and plugin index to match.
Validation: 97 focused scheduler, timezone, tool contract, and WebUI extension tests passed. Confirmed the live instance serves the updated labels.
Telegram advertised project/profile settings while its lifecycle and sender
authorization read only global configuration. A saved scoped allowlist or
disable could therefore leave a permissive global bot accepting messages.
Make Telegram configuration and activation explicitly global, preserving
per-user project routing. Enforce manifest scope support before plugin
configuration or activation writes and return HTTP 400 for invalid saves.
Pause Telegram bots and block incoming events whenever legacy scoped
configurations or activation overrides exist. Preserve those files, skip
attachment cleanup, notify the operator, and retain settings UI access for
manual review and removal. Each event must resolve a currently enabled
global bot; removed, renamed, disabled, or globally deactivated bots cannot
fall back to captured settings.
Add regressions for scope rejection without side effects, supported-plugin
scopes, legacy settings shutdown and migration, current allowlists, and
immediate revocation. Update existing administrator tests and owning docs.
Validation: 339 tests and 73 subtests passed in both Conda a0 and Docker's
framework runtime. A before/after reproduction showed an excluded sender
reaching context creation before the fix and being blocked afterward.
Restarted the live runtime and checked HTTP health and Telegram settings UI.
No live Telegram message test was performed.
input_remote forwards keyboard text as terminal code, allowing a custom
profile to bypass a blocked code_execution_remote capability through an
independently enabled input tool.
Require the canonical bundled remote terminal action in shared policy
resolution while retaining input_remote's independent permission. Apply
the dependency to execution, text prompts, and native tool schemas, and
recheck it inside the input wrapper before rewriting args or dispatching.
Output-only grants, default-deny policies, and profile tool-name overrides
cannot authorize the terminal operation.
Document the permission dependency and add 19 regressions covering zero
dispatch on denial, action-specific grants, prompt/schema visibility,
authorized input, host access controls, and framework-owned allow_running.
Validation: 232 focused tests passed in both Conda a0 and Docker's framework
runtime. A smoke test through the real CLI shell backend verified denied
dispatch, authorized idle and interactive input, and permission revocation.
Restarted the live Agent Zero runtime and verified HTTP health.
Cross-origin related windows could read page content and invoke DOM actions
through the helper's public postMessage channel.
Reject foreign and opaque origins, use exact target origins, and bind each
pending response to its child window and operation. Preserve inherited
origins for about:blank/srcdoc frames and trusted direct automation access.
Bump the helper to v2 and require a page reload when legacy listeners are
present. Existing host browsers must restart and reconnect before restored
tabs reload so old initialization scripts are discarded.
Add real-Chromium regressions for parent/frame/opener attacks, opaque
origins, forged responses, same-origin actions, and helper upgrades.
Validation: 250 browser/connector tests passed in both Conda a0 and Docker's
framework runtime; JavaScript syntax and diff checks passed. Restarted the
live runtime and verified HTTP health, browser navigation, and the viewer.
Accepted WhatsApp senders could persistently relax their own tool policy
and toggle instance-wide plugins using ordinary chat authorization.
Add a separate admin_numbers allowlist, empty by default, and derive
administrative access from the normalized sender of each inbound message.
Keep allowed-number and group checks in force. Require administration at
the WhatsApp plugin-toggle and tool-policy mutation boundaries while
preserving read-only menus and ordinary chat controls.
Gate built-in profile creation and browser-runtime changes before shared
command execution. Use the shared argument parser for profile creation
checks and preserve existing profile selection and custom overrides.
Expose the administrator list in settings and document the new defaults.
Add regressions for denied writes, real policy persistence and runtime
decisions, normalized sender identities, revocation, administrative
commands, and independent frontend access-list editing.
Validation:
- Conda framework: 165 tests and 62 subtests passed.
- Docker framework: 12 focused tests and 32 subtests passed.
- Node bridge and configuration store: all 4 tests passed.
- WebUI: administrator field renders, parses and clears independently.
- JavaScript syntax and git diff checks passed.
- Restarted the bound Docker instance and verified HTTP 200.
Live WhatsApp delivery was not exercised. Synthetic events and disposable
policy files cover the authorization paths without changing live access
lists or sending test messages to WhatsApp users.
Accepted Telegram senders could use /permissions and its callbacks to
persistently remove their own tool restrictions. The same access also
exposed instance-wide plugin toggles.
Add a per-bot admin_users allowlist and enforce it when opening settings
and processing every settings callback. Administrators must still pass
allowed_users; an omitted or empty admin_users list grants nobody settings
access. Resolve membership from the current bot configuration on each
event so revocation applies to existing buttons.
Expose the administrator list in the WebUI and document its defaults.
Add regression coverage for message and callback routes, denied writes,
authorized policy persistence, runtime tool decisions, and revocation.
Validation:
- Conda framework: 123 tests and 41 subtests passed.
- Docker framework: 9 focused tests and 41 subtests passed.
- WebUI: administrator field renders, parses entries, and clears correctly.
- JavaScript syntax and git diff checks passed.
- Restarted the bound Docker instance and verified HTTP 200.
Telegram network delivery was not exercised; tests use synthetic events
and disposable policy files without changing live bot access settings.
Restore matching command previews from the stored, masked command when
copying execution output or opening step details. Preserve command
whitespace and the output body while keeping the chat preview compact
and retaining existing log size limits.
Cover Bash, PowerShell, Python and Node.js prefixes, streamed updates,
masked placeholders and output without a matching command preview.
Validation: 48 focused WebUI tests passed, git diff --check passed, and
an actual browser clipboard check passed using frontend assets served
by the live instance at localhost:32081.
Update the default Codex client version from 0.165.0 to 0.166.0 for
upstream request headers and model catalog discovery. Keep the OAuth
plugin contract in sync; saved version overrides retain precedence.
Validation: 222 OAuth and plugin scan tests passed in the a0 framework
environment. Confirmed the bundled default in the live container;
git diff --check passed. Live upstream requests were not tested.
Preserve free text and reasoning-only output in the current generation
and retry without intermediate warning rows. Require an explicit response
tool call to finish in both API modes, including after provider built-in
actions.
Keep history metadata, Responses state, tool validation and the
consecutive-failure stop intact. Allow core retries when Context Doctor
and its prompts are unavailable.
Validation: 411 backend tests and 2 Chromium live/replay grouping tests
passed; live runtime health check passed.
Remove card descriptions and rename Tasks to Scheduled Tasks and Visit
Website to Visit Agent Zero Website.
Add a persisted half/full-width control. Narrow panels use square tiles
with labels below the icons, and desktop half-width remains available
without a neighboring half-width block. Update the owning documentation
and tests for independent width preferences.
Validation: 66 focused checks passed. Verified the live WebUI across
320-1920px viewports, both themes, reload persistence, keyboard controls,
unpaired half-width layout, and the Memory and Scheduled Tasks modals.
Display each channel name directly so entries read Telegram, Email,
WhatsApp, or the installed plugin name. Update the discovery contract and
existing regression assertion to match.
Validation: all six welcome composer tests pass, and the live runtime
serves the updated markup. Browser rendering was not rechecked.
Discover installed custom plugins carrying the exact channels Plugin
Index tag. Prefer their native thumbnail, fall back to the index image,
and open the existing plugin settings modal from each card.
Skip suspended or unlisted plugins, missing config screens or thumbnails,
and integrations that already contribute a discovery card. Preserve
bundled cards during background index loading and retain dismissal state.
Validation: 10 focused Node tests and 23 framework static tests passed.
Verified the live Discord A0 thumbnail and settings action at
localhost:32081 without saving configuration changes.
Add a puzzle-icon "Find channels plugins" text button below the channel
buttons, reusing the welcome footer styling. Open Plugin Hub Browse with
the channels tag selected and any previous search cleared. Keep the
Channels filter visible even when no plugins have the tag.
Update the owning docs and regression coverage. Verified with 21 focused
tests and live browser checks for navigation, search reset, mobile layout,
and keyboard focus.
Keep the move menu open when focus passes between its buttons so Move down
works with both mouse and keyboard.
Add a saved half/full-width toggle for System Resources and adapt its meters
and footer to the panel width. Stretch the accounts and channels cards to
match their row height, together or beside taller panels.
Keep card drag handles visible while preserving hover and keyboard-focus
visibility for the greeting/composer and Quick Actions controls.
Add width persistence coverage and update the owning UI contracts. Verified
with focused Node/Python tests and live browser checks for ordering, saved
layout, sizing, drag controls, desktop/mobile layouts, and both themes.
Suggest gpt-6.1-terra for Main and gpt-6-luna for Utility through Codex
provider metadata, while preserving explicit model lists and the live
account catalog.
Change the final onboarding action to Continue and close the wizard without
creating another chat. The existing model-gate event resumes the held message
in its original chat. Update the onboarding guide and ready-screen capture.
Add regressions for recommended and explicit model selection, configured
catalog order, held-message dispatch and completion without a pending chat.
Validation: 244 focused Python tests, three onboarding Node tests and the
OAuth reconnect check passed. Verified on the dedicated instance that
Continue closes the wizard and preserves the selected chat and chat list.
Label the welcome puzzle shortcut "Find welcome screen plugins" and give
both footer controls the same muted text and icon styling. Replace hover
backgrounds with a smooth 180 ms brightness transition, retain a visible
keyboard focus outline, and respect reduced motion. Let the controls wrap
on narrow screens.
Expand the onboarding shortcut to "Didn't find your provider? Click to
look in the Plugin Hub" across Cloud, Account, and Local.
Validation: 12 Node tests, 6 welcome UI tests, and 7 onboarding tests passed.
Verified the live hover transition and Welcome-filtered Plugin Hub.
Add neutral puzzle-icon links to the first-message setup prompt and all
three onboarding provider tabs. Open Plugin Hub with the exact provider
tag filter, clear previous searches, and keep Provider visible in the
filter list even when no plugins match.
Replace the Account and Local notes with the shared shortcut. Close the
wizard before opening the Hub and preserve the pending chat message.
Validation: 24 focused pytest tests and 4 Plugin Hub tests passed.
Verified all three tabs, keyboard activation, and the 14 tagged results
in the live UI.
Several test files still modeled older behavior: goal plugin stubs lacked the agent_name and context.log attributes the response tool reads, the intervention test built an AgentContext without id, startup asset assertions forbade the favicon link that safe and splash pages now ship, document query assertions pinned a superseded skill description, and the defer weakref check raced cyclic garbage collection. Stubs and assertions now match the code they cover. The welcome splash test asserted that the transparent background string appears nowhere in the welcome screen, which stopped holding once the reorderable layout menu gave its disabled buttons that background, so the check now inspects the welcome container block it was written for. The notification store test strips imports from the module it evaluates, which left the extension hook undefined once the store reported every received notification, so the harness now declares that hook.
Validation: full suite in a nested dev checkout fails 0, passes 3067, skips 7, down from 32 and 3029 at the first commit and 9 and 3058 after the path fix; pytest tests/test_webui_startup_assets.py tests/test_webui_message_ordering_static.py tests/test_ws_client_api_surface.py tests/test_defer_lifecycle.py tests/test_document_query_plugin.py tests/test_intervention_llm_abort.py plugins/_goal/tests/test_goal_plugin.py (67 passed).
A dev checkout nested under /a0 produced /a0-prefixed real paths that fix_dev_path stripped unconditionally and that time travel re-prefixed with the base dir, so workspace paths doubled into /a0/a0/usr/... and time travel reported missing workspaces or refused operations outside the selected scope. fix_dev_path now leaves paths that already live inside the base dir untouched, time travel resolves in-checkout display paths directly, and canonical workspace display paths are posix-normalized so the /a0 root keeps its stable workspace id. Because canonical display paths are now logical /a0 paths instead of physical nested checkout paths, workspace ids derived from them change once for dev instances nested under /a0, so their existing shadow repos are re-initialized in a new location and the old directories stay on disk unreachable; production instances whose base dir is /a0 keep their ids, and the time travel plugin DOX records that one-time change.
Validation: pytest tests/test_time_travel.py tests/test_dev_path_resolution.py (21 passed); full suite in a nested dev checkout fails 9, passes 3058, skips 7, down from 32 and 3029 at the previous commit, with the time travel and commands path cases now green; removing the nested-checkout guard, the separator normalization, or the canonical path change each fails its matching new case.
Walk through Pi, Hermes Agent and Claude Code exports, imports, separate
working-file uploads and continued tasks using real CLI sessions. Add
built-in chat and message search instructions and optimized neutral captures.
Preserve Claude custom titles and keep tool-result blocks in historical
tools instead of duplicating them as user messages. Retain unmatched results
and error markers, including coverage through the shared OpenClaw parser.
Validation: 28 migration tests and 11 subtests; six live imports into three
projects; three completed continuations; two booking tests; documentation
links, image privacy checks and rendered previews.
Accept an optional query on the protected chats_list route to search loaded chat titles and message logs. Reuse the bundled sidebar search helper for bounded, Unicode-safe literal highlight parts while keeping the normal list response unchanged.
Validate request and query types and limit queries to 200 characters. Blank queries skip scanning. Document the API contract and cover unchanged listing, authentication, validation, and excerpts.
Validation: 29 Core API and shared search-helper tests passed in the live Docker framework. Verified the live CLI client against 400 loaded conversations. Whitespace checks passed.
Search loaded chat and task titles and messages from the sidebar, with Standard, Compact, and No search modes in the existing filters menu. Preserve project, status, and activity filters, keep matching child conversations accessible, and leave Welcome Chats search independent.
Show compact, title-aligned excerpts with Unicode-safe match emphasis and no duplicate title previews. Use the Search placeholder and the chat composer's background, border, and focus transition. Retain authentication, query validation, stale-response protection, and keyboard navigation.
Validation: 33 focused Python tests and both frontend suites passed. Checked search modes, persistence, child/task fixtures, dark and light themes, keyboard focus, and touch emulation against the live Docker runtime. Final CSS polish was visually checked; git diff --check passed.
Keep reasoning effort, reasoning summary, and answer verbosity editable below the Codex quota before and after connection, including reconnects and selection of another provider. Reuse the existing controls and save behavior.
Use theme text and muted-text colors for provider details to prevent the connected-state green from leaking into settings. Document the visibility contract and cover connection, selection, and reconnect states in the existing JavaScript regression.
Validation: 221 OAuth and plugin-scan tests passed, along with the JavaScript regression and git diff --check. Verified the connected Codex panel and corrected colors in the live Web UI at localhost:32081.
Import Pi JSONL session versions 1–3 through the existing preview and category-selection workflow. Preserve separate retained branches, tool arguments and results, errors, public shell history, titles, and project context while excluding system prompts, reasoning, extension state, and image bytes. Validate session trees and bound branch expansion before processing.
Persist migration provenance in durable chat data, redact imported titles, and accept Unicode line separators inside JSONL strings. Add the Pi source selector and a user guide with two optimized screenshots from a real preview and import.
Validation: 25 migration tests and 11 subtests passed; the live Web UI imported two real Pi chats into one project with saved provenance and historical tool records; source-session hashes were unchanged and private reasoning was excluded. Checked desktop/mobile layout, 487 local documentation references, JavaScript syntax, image metadata/privacy, and git diff --check.
Replace the onboarding terminal preview and CLI guide screenshots with optimized full-colour captures of the current vanilla CLI. Use neutral workspace paths, remove image metadata, and keep shared captures identical across the documentation repositories.
Update host-picker wording, automatic connection guidance, chat tabs and model controls, and Browser mode scope. Version the onboarding image URL to refresh browser caches and document capture ownership.
Validation: reviewed onboarding at desktop and mobile sizes, verified documentation links and image integrity, and passed git diff --check.
Keep account and channel cards at their natural height when paired with a taller welcome plugin. Adapt account content to the card width and wrap channel buttons without stretching account badges.
Use the neutral text palette for the shared Files and Editor tree clear icon. Update the owning layout contracts.
Validation: 8 welcome Node checks and 8 focused Python tests passed. Live checks covered half/full width, narrow desktop and 390px mobile layouts, both themes, and file-tree filtering.
Add a persistent puzzle-piece button beside Show all dismissed that opens
Plugin Hub Browse with Welcome selected and any previous search cleared.
Preserve this requested filter through the Browse initialization reset.
Match and count all plugin tags rather than only the first tag, and keep
the Welcome filter visible even before matching plugins are indexed. Clip
the welcome background without creating a nested scroll container so block
focus and touch reordering scroll the page correctly.
Validation: focused Hub and restoration tests passed, along with six
welcome composer checks. Live verification covered the Welcome filter,
responsive plugin panels, touch reordering, dismissal/restoration, and
clean console output. Shared component and extension loaders are unchanged.
Allow welcome blocks to be reordered with mouse and touch drag handles,
keyboard controls, and a move menu. Persist the order in the browser,
commit DOM order on drop, and provide accessible announcements, reduced
motion, cancellation, and per-mount cleanup.
Discover blocks from their HTML metadata so plugins can join dynamically
without a central ID list. Preserve positions while plugins are absent,
validate unique IDs, and pair adjacent half-width blocks responsively.
Keep discovery cards in their Alpine scope and leave the shared component
and extension loaders unchanged.
Place the single Show all dismissed control below every welcome block and
extension contribution, retaining left alignment and unified restoration.
Document the plugin contract and cover ordering and restoration behavior.
Validation: 80 focused Node and pytest checks passed during implementation.
Verified mouse, emulated touch, keyboard, persistence, dynamic plugin
registration, and restoration in the live WebUI at localhost:32081.
Verified the final restore-button placement and reran both restore tests.
Allow the greeting/composer and Quick Actions to be hidden independently, with browser-local persistence and hover, keyboard, and touch-accessible close controls. Preserve the composer draft while hidden.
Match System Resources controls to discovery cards, retain readable light-theme colors, and replace the two restore paths with one flush-left Show all dismissed action for sections, notices, accounts, and channels.
Validation: eight focused tests passed. Verified desktop, touch, keyboard focus, both themes, draft preservation, partial/all-dismissed states, reload persistence, and restoration in the live WebUI at localhost:32081.
Update the default Codex client version from 0.160.1 to 0.165.0 for upstream request headers and model catalog discovery. Keep the OAuth plugin contract in sync; saved version overrides retain precedence.
Validation: 221 OAuth and plugin scan tests passed in the a0 framework environment; git diff --check passed.
Keep the original model response in turn-local repair data so Context Doctor
honors Update log content: off retains raw XML/DSML in View Details, while
on shows repaired JSON. Mark successful recovery notices as utility messages
so they follow the existing UTL visibility preference.
Accept an outer headline alongside DSML tool_args and preserve it through
normalization. Keep parallel calls, nested arguments and command escaping
intact, with normal repair fallback and dispatch unchanged.
Add regression coverage for both transports, log-setting states, disabled
Context Doctor, repeated repair hooks and outer versus nested headlines.
Update the owning documentation and include the reported DSML fixture.
Validation: 263 focused tests passed locally and in the Docker framework
runtime. The live instance restarted successfully and its health check passed.
Open Easy mode with only the main instruction prompt and fetch the remaining
prompt files on first Advanced entry. Preserve unsaved edits, checkpoints,
project scope changes and ACE undo history. Keep full loads as the API default
for existing clients.
Inventory prompt files once per root per request and render capability lists
and tool actions only when expanded, avoiding work while those views are closed.
Validation: 325 focused regression tests pass in the Docker framework runtime.
Live browser and HTTP checks cover Global/project saves, deferred loading,
mode switching, unsaved edits, action controls, ACE undo, and failed-load retry.
The full suite reports 2,939 passed and 48 skipped; all 14 failures also reproduce
against the unchanged HEAD before this change.
Let unconverted alternate-format output reach Context Doctor's existing JSON
repair path. Remove the leaked-call rejection state, retry hook and prompts
so missing delimiters and other repairable output can execute in the same
turn, restoring the behavior from before PR #1920.
Keep complete XML/DSML/native-text conversion, normal tool permissions,
Context Doctor's XML history cleanup and existing output-limit handling.
Update integration tests and ownership documentation for the restored flow.
Validation: 504 targeted tests passed both locally and in the Docker framework
runtime, including actual agent-loop tests for missing brackets and braces.
Historical comparisons matched pre-PR behavior in 48 cases and confirmed
12 additive format-conversion cases. Restarted the local runtime and verified
its health and extension discovery.
Separate leaked-call classification from retry handling so Context Doctor
can apply XML suppression, thought splitting, and log settings before the
history entry is saved. Keep rejected calls out of JSON repair and tool
dispatch; disabling XML suppression retains them only as thoughts.
Core still owns bounded retries and history insertion, defaults to {} when
the plugin is disabled, and clears replayable Responses output and pending
continuation state. Update plugin settings guidance and ownership docs.
Validation: 455 focused tests passed locally and in Docker's framework
runtime. Restarted the idle live app, verified health and actual extension
ordering, and found no startup error markers.
Recognize a standalone DSML calls/function_calls envelope containing one
invocation. Preserve explicitly string-valued parameters, strictly decode
JSON-valued parameters, and unwrap a sole object-valued tool_args field.
Keep canonical streaming parsing unchanged. Reject incomplete, malformed,
duplicate, mixed-wrapper, multiple-call, or prose-surrounded payloads;
quoted examples remain non-executable. Recovered calls still require an
offered tool and pass the normal execution-time permission checks.
Validation: 378 focused tests passed locally and in Docker's framework
runtime, including the reported DSML command as parser-only test data,
Chat Completions/Responses dispatch, and blocked-tool enforcement.
Store a neutral {} for rejected or incomplete leaked calls so retries do
not reinforce malformed XML or JSON. Keep the original result available
for diagnostics and preserve the existing warning and retry limits.
Remove replayable output items from the history copy and clear the
Responses continuation after normal history insertion. Retain recorded
turn IDs and usage while forcing the next request to use cleaned history.
This works with Context Doctor enabled or disabled.
Validation: 349 focused parser, dispatch, history, Responses, streaming,
Context Doctor, and retry-limit tests passed for this standalone change.
Show each tool once in Easy and Advanced modes, with expandable action
controls and existing descriptions. Whole-tool controls update all actions;
action searches reveal matching entries without narrowing that scope.
Hide permission summaries when every action is enabled. Add an expanded
"Individual tool actions" subtitle and align action text and controls with
the parent tool. Preserve saved action IDs and keep expansion and search
out of profile changes.
Validation: 194 focused tests passed, with all seven WebUI tests rerun
after the final styling changes. Live Docker checks covered desktop/mobile
rendering, keyboard expansion, filtered bulk edits, and save/reopen. The
final padding adjustment was verified in the served stylesheet.
Let profiles allow individual operations across all 12 bundled multi-action
tools, such as reading files without allowing writes or listing scheduled
tasks without allowing deletion.
Share action contracts across runtime policy, Agent Editor, text prompts,
and native schemas. Preserve existing whole-tool policies and legacy memory
IDs; only explicit permission edits persist the editor's projection.
Check aliases, defaults, browser batches, reset flags, and parallel
operations at execution. Apply action permissions to automatic browser
context and skill hints.
Validation: 742 focused tests passed. Live Docker checks verified allowed
reads, denied writes through native/legacy/parallel dispatch, blocked
scheduler deletion, and editor save/reload.
Add matching catalog artwork for A2A Chat, Behavior Adjustment,
Notify User, Scheduler, Search Engine, and Wait, completing thumbnail
coverage for all 47 bundled plugins.
Save each JPEG at 256x256 and below 20,000 bytes, and document the
asset ownership in each plugin's DOX.
Validation: checked image dimensions, encoding, and file sizes;
verified catalog discovery, HTTP delivery, and live WebUI rendering.
Initialize the scheduler store before snapshots. Preserve legacy memory calls and per-action permissions while advertising the unified tool, and retain saved policy IDs for moved tools. Restore legacy log rendering and connector tool events, and reject malformed boolean arguments by default.
Add a capability-gated, session-authenticated and CSRF-protected list/set API using the existing agent-scoped extension dispatcher. Plugins own descriptors, domain validation, revision checks and persistence; the connector backend has no optional-plugin adapters or arbitrary URL/code dispatch.
Document the extension and transport contract and cover independent controls, scoped availability, input validation and unchanged error responses. Validation: 21 focused Core tests passed; the live endpoint rejects missing CSRF and discovers the optional plugin control.
Prompt templates with runtime variables looked read-only and exposed processor
details above the editing area. Give main sections readable titles and show
customization status beside the title, a compact save path with a reliable
copy action, and a short variable hint below the editor.
Keep authored placeholders intact so runtime values refresh normally. Cover
Global and project overrides, template hints, search, and clipboard failures;
update the plugin documentation.
Validation: 198 focused tests passed. Verified save/reopen/reset, clipboard
contents, and a 377px-wide editor pane on the live localhost:32081 instance.
Framework warning log rows no longer repeat the agent name before the message text, covering tool and execution errors, the misformat notice, and the truncated, empty, repeat and fallback response warnings. A source-level regression test keeps the warning rows prefix-free, and the message-loop and context-doctor DOX and tests stop expecting the removed prefix.
Validation: pytest tests/test_warning_log_prefixes.py tests/test_tool_action_contracts.py tests/test_message_loop_result.py plugins/_context_doctor/tests/test_context_doctor.py
The computer use remote tool logs its messages as a dedicated computer_use_remote type with an icon verb headline naming the action, and a plugin message handler renders them as DSK badges in their own lilac accent shared with the agent-to-agent badge, with arguments passed through. Regression tests cover the log entries and headlines.
Validation: pytest tests/test_a0_connector_computer_use_tool_ui.py
The subordinate, unknown, MCP, and generic fallback tool log headlines no longer repeat the agent name. The unknown tool inherits the prefix-free base heading and takes the class-level blank line used by sibling tools, MCP keeps the tool name visible as an argument row, parallel children without a resolvable dedicated tool log through the prefix-free fallback, and the input and text editor fallback headings drop the prefix too. Framework helper DOX and both plugin AGENTS files document the prefix-free fallback rows, and regression tests cover the named headings for method and no-method calls, the generic fallback rows, and source guards for the plugin fallbacks.
Validation: pytest tests/test_tool_heading_prefixes.py tests/test_text_editor_message_handler.py tests/test_parallel_tool.py