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.
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.
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.
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.
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.
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.
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.
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.
Add Browser Settings guidance for supported Chromium-family remote debugging pages, including Opera and explicit remote-debugging endpoint fallback.
Update BYOB docs and troubleshooting to clarify that browser inventory comes from the connected A0 CLI and requires reconnecting the CLI after authorizing a browser.
Poll connected A0 CLI host-browser metadata while Browser Settings is open so newly authorized endpoints appear without saving or reopening.
Extend Browser Settings labels and remote-debugging guidance for Brave, Opera, Vivaldi, and Chromium-family host browsers with regression coverage.
Add host_browser_selection config normalization, Browser Settings UI choices from CLI-advertised inventory, and browser_selection forwarding through connector browser operations.
Expose host browser ids, labels, and available_browsers through A0 connector metadata and the browser_runtime API, with regression coverage for selection normalization.
Expose Browser settings for separate per-chat tabs versus a shared tab strip, and surface the existing maximum-tabs-per-chat cap. Make viewer WebSocket payloads carry the selected tab scope so the Browser panel can render the right tab list without guessing.
Keep the default scoped per chat, preserve the shared mode as an opt-in fallback, and cover both paths with Browser regression tests.
Notify screencast consumers when a stream is stopped so the WebSocket frame supervisor restarts after viewport-driven screencast resets.
Keep the active canvas as the frame/input surface and avoid replacing live screencasts with click snapshots, preserving hover, click, and wheel feedback.
Tested: conda run -n a0 pytest tests/test_browser_agent_regressions.py -q
Wake the WebSocket screencast supervisor when the runtime frame task stops before delivering a consumer frame, and keep fatal frame exits eligible for normal CDP teardown.
Avoid resizing the Browser canvas backing store on every frame when bitmap dimensions are unchanged, preventing unnecessary clears and reallocations.
Render live browser screencast frames through a canvas/ImageBitmap path while keeping the existing image/data-url path for snapshots and fallback rendering.
Attach WebSocket streams as runtime screencast consumers so delivered frames are pushed to the server loop and acknowledged after emit, removing the per-frame read_screencast_frame runtime round trip.
Keep the live browser frame proportional while the viewer is being resized by using contain sizing instead of stretching the stale frame to the transient surface shape.
Update the regression guard so the viewer contract pins the proportional render path.
Disable WebUI binary-frame negotiation until its Socket.IO client reconstructs frame attachments as real Blob, ArrayBuffer, or typed-array values.
Reject placeholder binary payloads instead of wrapping them into broken image blobs, preserving the slim-frame path through base64 frames.
Document the transport constraint and pin the fallback behavior in browser regression coverage.
Negotiate binary and slim screencast frames so updated clients avoid base64-in-JSON and repeated viewer metadata on every frame.
Clamp screencast capture to the subscriber viewport and DPR, lower streaming JPEG quality while preserving snapshot quality, and teach the client to manage binary frame URLs safely.
Document the browser viewer performance plan and update regression coverage for the new transport contract.
Replace the WebSocket viewer frame pump's idle poll with read_screencast_frame so live screencast delivery follows the producer cadence instead of a 50 ms sleep cycle.
Keep idle-page timeouts inside the stream loop to avoid screencast restart flashes, and update the Browser regression test to pin the blocking read path and timeout handling.
Set LibreOffice Writer as the Desktop MIME default for Markdown and plain text while keeping Agent Zero Editor as the secondary Open With association.
Update the Desktop DOX contract and extend the routing regression test to pin the Writer-first association order.
Limit the Browser URL-intent handler to web URL schemes so custom Agent Zero schemes fall through to their owning surfaces.
Document the scheme-ownership contract and cover the a0-editor misroute regression in the desktop/editor routing test.
Render settings secrets as masked text inputs instead of browser password fields. Add a static regression test so settings password controls do not return during future refactors.
Render the OAuth model provider select as unchosen when the stored model slot still points at a non-connected or non-OAuth provider. This avoids showing Codex/ChatGPT as selected while the model input remains disabled, keeping account-backed model selection explicit.
Move active What's New content into an empty slide list so the startup modal stays quiet until new cards are added. Keep manual Open available with an empty state, and update the static test plus DOX contract for dormant periods.
Remove memory-plugin assumptions from root and non-plugin DOX contracts so disabled plugins are not implied as available runtime behavior.
Keep the memory plugin documented only in its owning plugin subtree and mark it optional from the parent plugin index.
Set the bundled model config, Balance preset, and onboarding OpenRouter default to anthropic/claude-sonnet-5. Refresh the model config README example to match.
Stop seeding stale Codex OAuth model metadata so connected accounts use the upstream /models response instead of the retired gpt-5.2-codex default.
Use gpt-5.5 as the Codex proxy fallback model, omit the fake legacy client version when Codex CLI is unavailable, and remove the stale Copilot fallback entry.
Add focused OAuth regressions for the current default model, upstream model-list behavior, and Copilot fallback ordering.
Update first-run onboarding docs for the welcome screen, in-chat model gate, Cloud/AI account/Local provider paths, and explicit main and utility model choices.
Replace stale onboarding screenshots with optimized current assets and add gate, account, and local-provider captures.
Remove composer blocking so sends always create or enter a chat and route missing-model messages through the in-thread gate.
Make Utility model selection explicit by removing same-as-main sync and defaulting to each provider's utility model when available.
Compact onboarding success channel cards and update static regressions for the new onboarding/gate behavior.
Replace the separate Cloud/Local illustrated path screen with a single provider step that switches between Cloud, Account, and Local choices.
Route the first-send gate account choice through onboarding, remove the old Cloud/Local card assets, and keep model setup in the existing onboarding flow.
Update static onboarding and welcome composer regressions for the merged provider step.
Stop auto-populating Main and Utility models when an OAuth account is already connected or newly connected.
Add a connected-account gate state that routes users to model configuration instead of choosing provider defaults silently.
Update OAuth notifications, chat gate copy, DOX contracts, and focused static regressions for the explicit model-selection flow.
Move missing-model setup out of the welcome screen and into an in-thread gate that preserves the pending prompt, survives refresh, and auto-dispatches after a usable model is configured.
Reuse existing onboarding, OAuth account settings, and advanced model configuration surfaces instead of duplicating provider forms. Connected OAuth accounts can now populate Main and Utility defaults when no chat model is configured.
Update model readiness checks, focused static regressions, and DOX contracts for the new deferred setup flow.
Detect LiteLLM JSONDecodeError failures caused by the Responses mock streaming iterator trying to parse a real SSE stream.
Fall back to Chat Completions before any output is emitted, keeping custom OpenAI-compatible proxy prefixes provider-neutral.
Add a regression test for SSE-shaped JSON decode failures and update the transport DOX contract.
Verified with: PYTHONPATH="/home/eclypso/a0/agent-zero" conda run -n a0 pytest tests/test_stream_tool_early_stop.py tests/test_responses_architecture.py -q; PYTHONPATH="/home/eclypso/a0/agent-zero" conda run -n a0 python -m py_compile helpers/litellm_transport.py; docker exec -i -w /a0 9c436228a4a2 /opt/venv-a0/bin/python -m py_compile helpers/litellm_transport.py; git diff --check.
Mark message action button chrome as non-selectable so copy/paste captures message text without toolbar icon labels.
Add a static regression for the shared action-buttons stylesheet.
Render file action menus outside the scroll container with fixed positioning so dropdowns remain visible while file list scrolling stays intact.
Deduplicate component body assets during x-component loading and add regressions for the menu, opaque header, and duplicate scoped styles.
Only treat top-level JSON objects as tool roots during streaming, so a complete nested tool_calls item cannot end the stream before the parallel wrapper closes.
Add regression coverage for partial parallel wrapper snapshots.
Switch the chat composer to a Markdown-backed contenteditable editor so typing a triple-backtick fence and pressing Enter creates a visual code block while pasted fenced Markdown stays plain text.
Serialize visual code blocks back to fenced Markdown for send/history, keep full-screen input in sync with the backing store, and update the focused composer regression plus DOX contract.
Mark chats after successful persistence and reconcile saved chat IDs while building WebUI state snapshots so stale in-memory contexts disappear after their chat files are removed.
Keep fresh unsaved chats visible, skip running contexts, and cover the regression with snapshot and persistence tests.
Fall back to Chat Completions for provider/proxy Responses endpoint failures before any output is emitted.
Preserve streamed Chat Completions tool-call deltas as structured LLMResult function-call items, so fallback providers can still drive tools.
Document Docker host-gateway addressing for local model servers, where container localhost does not reach host loopback.
Verified with: PYTHONPATH="/home/eclypso/a0/agent-zero" conda run -n a0 pytest tests/test_stream_tool_early_stop.py tests/test_responses_architecture.py -q; PYTHONPATH="/home/eclypso/a0/agent-zero" conda run -n a0 python -m py_compile helpers/litellm_transport.py helpers/llm_result.py; git diff --check.
Normalize memory_load threshold and limit values from model calls before vector search so numeric strings from Responses tool arguments do not reach FAISS unchanged.
Add a focused regression covering string threshold/limit values and document the memory plugin contract.
Notify users when automatic chat renaming cannot reach the Utility Model, while keeping the rename task best-effort. Normalize saved generated titles, mark WebUI state dirty after successful saves, and cover the success and failure paths with focused tests.
Refresh the root README download table from A0 Launcher v0.9 to v1.0 asset URLs and point the release-note link at the v1.0 release page. This commit intentionally sits after the v2.1 tag without moving the tag.
When Responses mode accepts a plain-text final answer, reuse the active generating log item as a finished response entry so the WebUI does not remain on the Calling LLM step.
Skip normal tool JSON and existing live response-tool logs, and cover the fallback with focused regression tests.
Ensure function parameter schemas include an explicit properties object before Responses requests are dispatched through LiteLLM.
Keep prompt/MCP permissive schemas compatible with stricter OpenAI-compatible chat validators and cover chat, legacy function, and native Responses tool inputs with regressions.
Treat full log snapshots as authoritative replays by clearing the message DOM when the snapshot starts at log no 0, then render logs in backend order. This prevents streamed responses from staying above earlier welcome/user messages after sync races while keeping incremental updates patch-based.
Let backup creation and restore cleanup use unlimited pattern scans so archives for agents with more than 50,000 files are complete. Keep UI preview and dry-run paths bounded for responsiveness while making the dry-run truncated flag safe for optional limits.
Add regression coverage for unlimited scans, backup creation, clean-before-restore, and restoring an archive entry past the old 50,000-file boundary.
Restore parallel normalization tolerance for provider/model outputs that stringify the tool_calls array inside tool_args.
The wrapper still expects normal tool-call objects after decoding, so nested parallel and document_query guards continue to run through the same validation path.
