fix(copilot): bundle of chat stream stability fixes (PK dedup, race, compaction, errors, chips)

[majdyz]

Why

Bundled bug-fix PR for the copilot chat stream surface. Multiple user-reported regressions on dev (chat-mode-option LD flag) — duplicate sends persisting two user rows on the same session, "prompt too long" recurring on long sessions (SENTRY-1207), assistant turn double-error UI, lost queued chips on the in-flight poll, SDK-resume rescuing context but the post-turn upload throwing it away, SDK-mode worker crashes on lazy-init lock race, compaction failures sending the same too-long payload back into the retry loop. All share the chat stream code path so one PR.

What

Atomic dedup at Postgres ChatMessage.id

The duplicate-send loophole (RMQ redelivery, browser/CDN retries, refresh+retype) is closed at the database layer:

• Frontend transport (prepareSendMessagesRequest) generates crypto.randomUUID() per logical send — stable across SDK-internal retries because the prepared body is reused.
• Backend StreamChatRequest.message_id becomes ChatMessage.id on insert.
• Postgres' PK uniqueness constraint catches duplicate inserts. append_and_save_message distinguishes ChatMessage_pkey (dedup signal → return None, route subscribes to existing turn) from ChatMessage_sessionId_sequence_key (sequence race, retried internally) from other failures.
• Optimistic in-memory append is rolled back on any save-failure path, not just PK collision.
• No new column, no Redis claim store — the existing @id @default(uuid()) PK is the atomic primitive.

Race-safe HTTP handler

Dropped the redundant second is_turn_in_flight check in routes.py that introduced a TOCTOU window where a concurrent turn could leave a user message saved to the DB but never enqueued. The first check at the top of the handler already routes the in-flight branch to queue_pending_for_http; anything past that point starts a fresh turn.

Race-safe queued-message chips

useCopilotPendingChips chips now carry frontend-only UUIDs ({id, text}[] instead of indexed string[]). Mid-turn poll's drain promotion uses a functional updater that filters by id, so chips enqueued during the in-flight getV2GetPendingMessages GET aren't overwritten by the stale snapshot's setState. One bubble per chip, preserving identity for the useHydrateOnStreamEnd substring match.

Persist context-error retry recovery

T2+ retry in sdk/service.py was dropping session_id to dodge "Session ID already in use", so the recovery CLI wrote to a random path while the post-turn upload silently grabbed the stale pre-failure file at the predictable cli_session_path. The rescued (compacted) transcript was thrown away every time, and the next turn --resumed the same bloated GCS copy. New helper delete_stale_cli_session_file clears the local file before the retry; session_id is preserved so the recovery write lands on the predictable path.

Compression-failure fallback

When _compress_messages fails (LLM summarize + truncate fallback both error), return [], True (drop history, mark compacted) instead of the originals. The originals would guarantee another Prompt is too long on retry — burning the retry budget for zero progress. Bare current message is the tightest possible compression without an LLM.

Dedup error UI

Backend appends a COPILOT_ERROR_PREFIX marker to session.messages AND yields a StreamError SSE event on every final-failure path. Frontend rendered both — same failure.display_msg, twice. New top-level lastAssistantHasErrorMarker memo gates the trailing red banner.

Empty-tool-call circuit breaker exclusion

No-arg tools (e.g. get_agent_building_guide) were tripping the breaker because their tool-call payload is genuinely empty. Excluded via _no_arg_tool_names().

Executor lock + pickle fixes

• CoPilotExecutor was raising TypeError: cannot pickle '_thread.lock' on forkserver start. Lock is now lazy-property-backed so it's not bound to the parent process state.
• The lazy-property pattern then introduced a TOCTOU race that Sentry caught — fixed by materialising the lock pre-fork in run(), before workers spawn.
• Foreground execution (no detached process) is restored so the helm chart's terminationGracePeriodSeconds applies cleanly to in-flight turns.

SSR-safe persist storage

copilotStreamStore persist middleware factory now returns a no-op Storage stub when window is undefined (Next.js SSR / vitest), instead of undefined!. Browser path is unchanged: still uses window.sessionStorage.

How

• idempotency_key field on StreamChatRequest was the original Stripe-style design but reverted to using ChatMessage.id directly — Postgres PK is the simpler authoritative dedup.
• Transport-tier integration test (copilotStreamTransport.test.ts) added to plug the gap that allowed the AI SDK messageId regression (replace-mode semantics, broke optimistic render) to slip past unit tests.
• delete_stale_cli_session_file reuses the same projects_base() traversal guard as read_cli_session_from_disk.
• useCopilotPendingChips keeps its public API (queuedMessages: string[], appendChip(text)) — only internal state shape changed.

Test plan

• Backend pytest — model_test, routes_test, executor (utils, processor, manager), sdk (service_helpers, retry_scenarios, prompt_too_long, session_persistence, context_fallback): 154+216 passed
• Frontend vitest — useCopilotPendingChips, ChatMessagesContainer, useSendMessage, copilotStreamTransport (new), copilotStreamStore: all green
• pnpm types, pnpm lint, pnpm format clean
• poetry run ruff format + black on changed backend files clean
• /pr-test --fix locally (native), 2 independent runs on different HEADs: PASS — concurrent identical-message_id POSTs subscribe-only, distinct clicks fresh turn, DB row counts match
• /pr-test on dev preview: PASS — 5 scenarios with screenshots, comment 4353117872
• Sentry threads addressed: 4 fixed (TOCTOU race HIGH, SSR storage HIGH, optimistic-pop MEDIUM, ack-on-success), 1 documented false positive

[coderabbitai]

Note

Reviews paused

It looks like this branch is under active development. To avoid overwhelming you with review comments due to an influx of new commits, CodeRabbit has automatically paused this review. You can configure this behavior by changing the reviews.auto_review.auto_pause_after_reviewed_commits setting.

Use the following commands to manage reviews:

• @coderabbitai resume to resume automatic reviews.
• @coderabbitai review to trigger a single review.

Use the checkboxes below for quick actions:

• ▶️ Resume reviews
• 🔍 Trigger review

Walkthrough

Adds a guarded helper to delete deterministic local CLI session JSONL and integrates it into SDK restore/retry logic; frontend changes convert pending-chip state to chips with stable ids, promote one bubble per chip, and suppress duplicate trailing error banners when inline error markers exist.

Changes

Cohort / File(s)	Summary
Backend SDK Service autogpt_platform/backend/backend/copilot/sdk/service.py, autogpt_platform/backend/backend/copilot/sdk/service_helpers_test.py	Adds delete_stale_cli_session_file(sdk_cwd, session_id, log_prefix) with a projects_base() traversal guard and best-effort unlink. Replaces inline exists/unlink logic in _restore_cli_session_for_turn and ensures non---resume retries preserve session_id and proactively delete predictable stale session files. Tests added for deletion behavior and session-id selection.
Frontend Pending Chips Tracking autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts, autogpt_platform/frontend/src/app/(platform)/copilot/__tests__/useCopilotPendingChips.test.ts	Replaces string queue with Chip[] ({id, text}); queuedMessages is derived from chip texts. Peek/polling are made session-race-safe. Auto-continue and mid-turn drains promote one bubble per chip (deterministic ids) and remove only promoted/drained chips by id. Tests updated/expanded.
Frontend Promoted Bubble Helpers autogpt_platform/frontend/src/app/(platform)/copilot/helpers/makePromotedBubble.ts, autogpt_platform/frontend/src/app/(platform)/copilot/helpers/__tests__/makePromotedBubble.test.ts	Changes makePromotedUserBubble signature to accept a single text: string (no multi-chip join). Documentation and tests updated; id-generation retained.
Frontend Error Banner Deduplication autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatMessagesContainer/ChatMessagesContainer.tsx, autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatMessagesContainer/__tests__/ChatMessagesContainer.test.tsx	Adds parsing of special inline markers (error / retryable_error) in assistant message parts and suppresses the trailing error banner when the last assistant message already contains such a marker. Tests adjusted to verify deduplication.

Sequence Diagram(s)

(omitted — changes are localized helpers, retry behavior, and UI logic; no new multi-component sequential flow requiring a diagram)

Estimated code review effort

🎯 3 (Moderate) | ⏱️ ~25 minutes

Possibly related PRs

• fix(backend/copilot): fix T2+ --resume by using CLI native session file #12777: Modifies CLI session restore/upload and resume/retry behavior in the same SDK area — strong overlap.
• fix(copilot): prevent duplicate side effects from double-submit and stale-cache race #12660: Touches stream_chat_completion_sdk and retry/session-file handling in service.py — related changes.
• fix(backend/copilot): sync TranscriptBuilder with CLI on mid-stream compaction #12401: Adds session transcript helpers and modifies session-file lifecycle in the backend — related to deletion/upload logic.

Suggested reviewers

• Swiftyos
• Bentlybro
• Pwuts

"I nibbled stale files late at night, so sessions may resume just right,
Chips arrive with names and hop into line,
One bubble per chip now blooms each time,
Silent errors that echoed gently cease,
The rabbit scuffs the path and tidies peace." 🐇✨

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 77.27% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Title check	✅ Passed	The title directly relates to the main changes: atomic dedup, race fixes, message-chip compaction, error UI dedup, and SDK context-recovery retry — all stabilization fixes for the copilot chat stream.
Description check	✅ Passed	The description comprehensively covers the changeset: it explains the rationale (user-reported regressions), documents each major fix (dedup, race-safety, chips, SDK retry, compression, error UI, tool calls, executor, storage), and provides test results.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/copilot-stream-errors-and-queue-bubbles

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[github-actions]

🔍 PR Overlap Detection

This check compares your PR against all other open PRs targeting the same branch to detect potential merge conflicts early.

🔴 Merge Conflicts Detected

The following PRs have been tested and will have merge conflicts if merged after this PR. Consider coordinating with the authors.

• Persist stable copilot message IDs through hydration #12676 (rotempasharel1 · updated 4d ago)

  • 📁 autogpt_platform/backend/backend/copilot/
    • model.py (2 conflicts, ~19 lines)
    • sdk/service.py (1 conflict, ~54 lines)

• feat(platform): add first-class org/workspace support — schema, auth, APIs, migration, frontend #12670 (ntindle · updated 2d ago)

• feat(platform): estimate CoPilot turn cost and require approval for high-cost requests #12877 (Rushi-Balapure · updated 8d ago)

• feat(platform/copilot): Reduce time to first output #12828 (Pwuts · updated 13d ago)

🟢 Low Risk — File Overlap Only

These PRs touch the same files but different sections (click to expand)

• fix(backend/credit): reserve TOP_UP for real Stripe checkouts; refunds use grant_credits #12959 (majdyz · updated 3m ago)

  • Shared files: autogpt_platform/backend/backend/api/features/chat/routes.py

• dx(frontend): migrate direct hook tests to component-level tests #12960 (0ubbe · updated 1h ago)

  • Shared files: autogpt_platform/frontend/src/app/(platform)/copilot/__tests__/useCopilotPage.test.ts, autogpt_platform/frontend/src/app/(platform)/build/components/BuilderChatPanel/__tests__/useBuilderChatPanel.test.ts, autogpt_platform/frontend/src/app/(platform)/copilot/__tests__/useCopilotPendingChips.test.ts

• feat(copilot): add goal decomposition step before agent building #12731 (anvyle · updated 7h ago)

  • Shared files: autogpt_platform/backend/backend/api/features/chat/routes.py

• feat(platform): Trigger On Anything #12740 (Pwuts · updated 1d ago)

  • Shared files: autogpt_platform/backend/backend/api/features/chat/routes.py

• feat(platform): redesign wallet tasks to be autopilot-centric #12722 (kcze · updated 2d ago)

  • Shared files: autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPage.ts

• fix(frontend/builder): persist agent name on Agent Executor block after reload #12372 (okxint · updated 3d ago)

  • Shared files: autogpt_platform/backend/backend/copilot/model.py

• fix(copilot): mandate gh auth status check before connect_integration #12852 (tianhaocui · updated 8d ago)

  • Shared files: autogpt_platform/backend/backend/copilot/sdk/service.py

• fix(backend/copilot): reuse existing credentials across chat sessions #12767 (0ubbe · updated 11d ago)

  • Shared files: autogpt_platform/backend/backend/copilot/baseline/service.py, autogpt_platform/backend/backend/copilot/sdk/service.py

Summary: 4 conflict(s), 0 medium risk, 8 low risk (out of 12 PRs with file overlap)

---

Auto-generated on push. Ignores: openapi.json, lock files.

[coderabbitai]

Actionable comments posted: 1

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In `@autogpt_platform/backend/backend/copilot/sdk/service.py`:
- Around line 1307-1340: The delete_stale_cli_session_file function has a TOCTOU
due to calling Path.exists() before unlink and logs full path on unlink errors;
remove the exists() check and perform the unlink directly after validating the
real_path prefix (using cli_session_path and projects_base as currently done),
catch FileNotFoundError and return False, and catch other OSError exceptions but
log only os.path.basename(real_path) (not the full path) in the error message
before returning False.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 93e33973-968d-4ed6-9ac1-fa7806ae19c7

📥 Commits

Reviewing files that changed from the base of the PR and between 4a1741c and 0dcd25f.

📒 Files selected for processing (8)

• autogpt_platform/backend/backend/copilot/sdk/service.py
• autogpt_platform/backend/backend/copilot/sdk/service_helpers_test.py
• autogpt_platform/frontend/src/app/(platform)/copilot/__tests__/useCopilotPendingChips.test.ts
• autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatMessagesContainer/ChatMessagesContainer.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatMessagesContainer/__tests__/ChatMessagesContainer.test.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/helpers/__tests__/makePromotedBubble.test.ts
• autogpt_platform/frontend/src/app/(platform)/copilot/helpers/makePromotedBubble.ts
• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

[codecov]

Codecov Report

❌ Patch coverage is 82.59494% with 55 lines in your changes missing coverage. Please review.
✅ Project coverage is 69.64%. Comparing base (6ead5a2) to head (cafa596).
⚠️ Report is 1 commits behind head on dev.

Additional details and impacted files

@@            Coverage Diff             @@
##              dev   #12948      +/-   ##
==========================================
+ Coverage   69.62%   69.64%   +0.02%     
==========================================
  Files        2135     2135              
  Lines      157921   158133     +212     
  Branches    16312    16322      +10     
==========================================
+ Hits       109946   110131     +185     
- Misses      44727    44745      +18     
- Partials     3248     3257       +9     

Flag	Coverage Δ
platform-backend	78.62% <82.22%> (+0.01%)	⬆️
platform-frontend	31.82% <82.96%> (+0.20%)	⬆️
platform-frontend-e2e	30.53% <30.43%> (-0.16%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Components	Coverage Δ
Platform Backend	78.62% <82.22%> (+0.01%)	⬆️
Platform Frontend	37.95% <83.08%> (+0.10%)	⬆️
AutoGPT Libs	∅ <ø> (∅)
Classic AutoGPT	28.43% <ø> (ø)

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[coderabbitai]

Actionable comments posted: 1

Caution

Some comments are outside the diff and can’t be posted inline due to platform limitations.

⚠️ Outside diff range comments (1)

autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts (1)

128-146: ⚠️ Potential issue | 🟠 Major | ⚡ Quick win

Guard against stale peek responses after session switches.

A response started for an old sessionId can still resolve and call setChips, causing old-session chips to reappear in the new session. Add a stale-response guard before mutating state.

Suggested fix

   useEffect(() => {
@@
-    void getV2GetPendingMessages(sessionId).then((res) => {
+    const requestSessionId = sessionId;
+    let cancelled = false;
+
+    void getV2GetPendingMessages(sessionId).then((res) => {
+      if (cancelled) return;
+      if (prevSessionIdRef.current !== requestSessionId) return;
       if (res.status !== 200) return;
@@
       setChips(() =>
         res.data.count > 0
@@
       );
     });
+    return () => {
+      cancelled = true;
+    };
   }, [sessionId, status, setChips]);

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In
`@autogpt_platform/frontend/src/app/`(platform)/copilot/useCopilotPendingChips.ts
around lines 128 - 146, The promise handler for
getV2GetPendingMessages(sessionId) can mutate state after the user has switched
sessions; capture the sessionId at request time (e.g., const callSessionId =
sessionId) and before any setChips or other state changes verify that the
current sessionId still equals callSessionId (or return early if it doesn't).
Apply this stale-response guard inside the then callback before the
turnStarting/sessionChanged logic so setChips is only called for the matching
session; reference getV2GetPendingMessages, sessionId, setChips, turnStarting,
and sessionChanged when locating where to add the check.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In
`@autogpt_platform/frontend/src/app/`(platform)/copilot/useCopilotPendingChips.ts:
- Around line 233-237: The bubble ID generation is inconsistent: auto-continue
uses `${assistantId}-${chip.id}` while mid-turn uses `chip.id`, causing
duplicate filtering to miss matches; update the calls to makePromotedUserBubble
(e.g., the call that currently passes `chip.id` around mid-turn and the call at
auto-continue) to use a single deterministic ID format such as
`${assistantId}-${chip.id}` so both promotion paths produce the same bubble ID
for the same chip and duplicate filtering works correctly.

---

Outside diff comments:
In
`@autogpt_platform/frontend/src/app/`(platform)/copilot/useCopilotPendingChips.ts:
- Around line 128-146: The promise handler for
getV2GetPendingMessages(sessionId) can mutate state after the user has switched
sessions; capture the sessionId at request time (e.g., const callSessionId =
sessionId) and before any setChips or other state changes verify that the
current sessionId still equals callSessionId (or return early if it doesn't).
Apply this stale-response guard inside the then callback before the
turnStarting/sessionChanged logic so setChips is only called for the matching
session; reference getV2GetPendingMessages, sessionId, setChips, turnStarting,
and sessionChanged when locating where to add the check.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: e87b16b1-c686-4e4d-91c2-015430e4e74a

📥 Commits

Reviewing files that changed from the base of the PR and between 0dcd25f and 9e8622c.

📒 Files selected for processing (2)

• autogpt_platform/backend/backend/copilot/sdk/service.py
• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• autogpt_platform/backend/backend/copilot/sdk/service.py

📜 Review details

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (18)

• GitHub Check: lint
• GitHub Check: integration_test
• GitHub Check: check API types
• GitHub Check: test (3.12)
• GitHub Check: type-check (3.13)
• GitHub Check: test (3.13)
• GitHub Check: type-check (3.12)
• GitHub Check: type-check (3.11)
• GitHub Check: lint
• GitHub Check: test (3.11)
• GitHub Check: Seer Code Review
• GitHub Check: check-overlaps
• GitHub Check: types
• GitHub Check: lint
• GitHub Check: end-to-end tests
• GitHub Check: Check PR Status
• GitHub Check: Analyze (python)
• GitHub Check: Analyze (typescript)

🧰 Additional context used

📓 Path-based instructions (8)

autogpt_platform/frontend/**/*.{ts,tsx,js,jsx}

📄 CodeRabbit inference engine (.github/copilot-instructions.md)

autogpt_platform/frontend/**/*.{ts,tsx,js,jsx}: Use Node.js 21+ with pnpm package manager for frontend development
Always run 'pnpm format' for formatting and linting code in frontend development

Format frontend code using pnpm format

autogpt_platform/frontend/**/*.{ts,tsx,js,jsx}: Fully capitalize acronyms in symbols, e.g. graphID, useBackendAPI
No linter suppressors (// @ts-ignore``, // eslint-disable) — fix the actual issue

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

autogpt_platform/frontend/**/*.{tsx,ts}

📄 CodeRabbit inference engine (.github/copilot-instructions.md)

autogpt_platform/frontend/**/*.{tsx,ts}: Use function declarations for components and handlers (not arrow functions) in React components
Only use arrow functions for small inline lambdas (map, filter, etc.) in React components
Use PascalCase for component names and camelCase with 'use' prefix for hook names in React
Use Tailwind CSS utilities only for styling in frontend components
Use design system components from 'src/components/' (atoms, molecules, organisms) in frontend development
Never use 'src/components/legacy/' in frontend code
Only use Phosphor Icons (@phosphor-icons/react) for icons in frontend components
Use generated API hooks from '@/app/api/generated/endpoints/' instead of deprecated 'BackendAPI' or 'src/lib/autogpt-server-api/'
Use React Query for server state (via generated hooks) in frontend development
Default to client components ('use client') in Next.js; only use server components for SEO or extreme TTFB needs
Use '' component for rendering errors in frontend UI; use toast notifications for mutation errors; use 'Sentry.captureException()' for manual exceptions
Separate render logic from data/behavior in React components; keep comments minimal (code should be self-documenting)

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

autogpt_platform/frontend/**/*.{ts,tsx}

📄 CodeRabbit inference engine (.github/copilot-instructions.md)

autogpt_platform/frontend/**/*.{ts,tsx}: No barrel files or 'index.ts' re-exports in frontend code
Regenerate API hooks with 'pnpm generate:api' after backend OpenAPI spec changes in frontend development

autogpt_platform/frontend/**/*.{ts,tsx}: Use function declarations (not arrow functions) for components/handlers
No any types unless the value genuinely can be anything
Keep render functions and hooks under ~50 lines; extract named helpers or sub-components when they grow longer

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

autogpt_platform/frontend/src/**/*.{ts,tsx}

📄 CodeRabbit inference engine (AGENTS.md)

autogpt_platform/frontend/src/**/*.{ts,tsx}: Use generated API hooks from @/app/api/__generated__/endpoints/ following the pattern use{Method}{Version}{OperationName}, and regenerate with pnpm generate:api
Separate render logic from business logic using component.tsx + useComponent.ts + helpers.ts pattern, colocate state when possible and avoid creating large components, use sub-components in local /components folder
Use function declarations for components and handlers, use arrow functions only for callbacks
Do not use useCallback or useMemo unless asked to optimise a given function

autogpt_platform/frontend/src/**/*.{ts,tsx}: Keep files under ~200 lines; extract sub-components or hooks into their own files when a file grows beyond this
Use generated API hooks from @/app/api/__generated__/endpoints/ with pattern use{Method}{Version}{OperationName}
Always import the -Icon-suffixed alias from @phosphor-icons/react (e.g. TrashIcon, PlusIcon, SquareIcon) — bare exports are deprecated
Do not use useCallback or useMemo unless asked to optimize a given function
Never use src/components/__legacy__/* — use design system components from src/components/

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

autogpt_platform/frontend/**/*.ts

📄 CodeRabbit inference engine (AGENTS.md)

No barrel files or index.ts re-exports in the frontend

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

autogpt_platform/frontend/src/**/*.ts

📄 CodeRabbit inference engine (AGENTS.md)

Do not type hook returns, let Typescript infer as much as possible

autogpt_platform/frontend/src/**/*.ts: Extract component logic into custom hooks grouped by concern, not by component, with each hook in its own .ts file
Do not type hook returns; let TypeScript infer as much as possible

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

autogpt_platform/**/*.{ts,tsx}

📄 CodeRabbit inference engine (AGENTS.md)

Never type with any, if no types available use unknown

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

autogpt_platform/frontend/src/**/*.{ts,tsx,js,jsx}

📄 CodeRabbit inference engine (autogpt_platform/frontend/AGENTS.md)

Avoid index and barrel files

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

🧠 Learnings (10)

📓 Common learnings

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12773
File: autogpt_platform/backend/backend/copilot/pending_messages.py:52-64
Timestamp: 2026-04-14T14:36:25.545Z
Learning: In `autogpt_platform/backend/backend/copilot` (PR `#12773`, commit d7bced0c6): when draining pending messages into `session.messages`, each message's text is sanitized via `strip_user_context_tags` before persistence to prevent user-controlled `<user_context>` injection from bypassing the trusted server-side context prefix. Additionally, if `upsert_chat_session` fails after draining, the drained `PendingMessage` objects are requeued back to Redis to avoid silent message loss. Do NOT flag the drain-then-requeue pattern as redundant — it is the intentional failure-resilience strategy for the pending buffer.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12445
File: autogpt_platform/backend/backend/copilot/sdk/service.py:1071-1072
Timestamp: 2026-03-17T06:48:26.471Z
Learning: In Significant-Gravitas/AutoGPT (autogpt_platform), the AI SDK enforces `z.strictObject({type, errorText})` on SSE `StreamError` responses, so additional fields like `retryable: bool` cannot be added to `StreamError` or serialized via `to_sse()`. Instead, retry signaling for transient Anthropic API errors is done via the `COPILOT_RETRYABLE_ERROR_PREFIX` constant prepended to persisted session messages (in `ChatMessage.content`). The frontend detects retryable errors by checking `markerType === "retryable_error"` from `parseSpecialMarkers()` — no SSE schema changes and no string matching on error text. This pattern was established in PR `#12445`, commit 64d82797b.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 0
File: :0-0
Timestamp: 2026-04-25T02:53:53.964Z
Learning: In `autogpt_platform/backend/backend/copilot/sdk/service.py` (PR `#12918`, commit 6576bf561):
- `_flush_unresolved_tool_calls` was renamed to `flush_unresolved_tool_calls` (public); all call sites updated, `# noqa: SLF001` suppressor removed.
- `_flush_orphan_tool_uses_to_session` and `_InterruptedAttempt.finalize` both return `list[StreamBaseResponse]`; the post-loop caller yields those events directly to avoid double-flush and skipped UI cleanup events.
- The three former post-loop blocks (partial restore + redundant re-flush + two separate `yield StreamError` sites) are collapsed into a single block driven by `_classify_final_failure` returning a `_FinalFailure(display_msg, code, retryable)` dataclass, so history marker and SSE yield share one source of truth.
Do NOT flag double-flush risk or mismatched history/SSE marker as issues in the post-loop section of `stream_chat_completion_sdk`.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12797
File: autogpt_platform/backend/backend/copilot/sdk/service.py:1991-2021
Timestamp: 2026-04-15T13:44:34.273Z
Learning: In `autogpt_platform/backend/backend/copilot/sdk/service.py` (`_run_stream_attempt`), the pre-create block (PR `#12797`) intentionally does NOT call `state.transcript_builder.append_assistant(...)` when inserting the empty assistant placeholder into `ctx.session.messages`. The transcript is left ending at the `tool_result` entry (N entries) while `message_count` metadata is N+1. This mismatch is benign and deliberate: on the next `--resume`, the SDK sees the transcript ending at `tool_result` and correctly regenerates the assistant response. Pre-appending the assistant turn to the transcript would suppress regeneration while leaving `session.messages[-1].content = ""` permanently (worse outcome). On the gap-fallback path, `transcript_msg_count (N+1) >= msg_count-1 (N)` means no gap is injected for the empty placeholder, which is correct because injecting an empty assistant message as context would mislead the SDK. Do NOT flag this transcript/message_count discrepancy as a bug.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12796
File: autogpt_platform/backend/backend/api/features/chat/routes.py:504-527
Timestamp: 2026-04-16T12:33:44.990Z
Learning: In `autogpt_platform/backend/backend/api/features/chat/routes.py`, `get_session` (PR `#12796`, commit 3771bfad9c1) closes the TOCTOU race between the initial `stream_registry.get_active_session()` pre-check and `get_chat_messages_paginated()` with a post-check re-verification: after the DB fetch, if `is_initial_load and active_session is not None`, it calls `get_active_session` a second time; if `post_active is None` (stream completed during the window), it resets `from_start=True`, `forward_paginated=True`, and re-fetches messages from sequence 0. Do NOT flag the double `get_active_session` call pattern as redundant — it is the intentional TOCTOU mitigation for pagination direction selection.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 0
File: :0-0
Timestamp: 2026-04-28T03:31:29.696Z
Learning: In Significant-Gravitas/AutoGPT PR `#12933` (`fix/stripe-checkout-link-auth-loop`), the initial approach of pinning `payment_method_types=["card"]` in `top_up_intent` and `create_subscription_checkout` (in `autogpt_platform/backend/backend/data/credit.py`) was reverted in commit `584b43a71` as it patched a symptom. The true root cause was in `update_subscription_tier()` in `v1.py`: a `current_tier_price_id is not None` guard was gating admin-granted DB-tier flips and short-circuiting them when the BUSINESS tier was pruned from the price-id LaunchDarkly flag. Do NOT flag `payment_method_types` absence in these checkout helpers as a Stripe Link bypass issue; the fix lives in the subscription tier update guard logic.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12356
File: autogpt_platform/backend/backend/copilot/constants.py:9-12
Timestamp: 2026-03-10T08:39:22.025Z
Learning: In Significant-Gravitas/AutoGPT PR `#12356`, the `COPILOT_SYNTHETIC_ID_PREFIX = "copilot-"` check in `create_auto_approval_record` (human_review.py) is intentional and safe. The `graph_exec_id` passed to this function comes from server-side `PendingHumanReview` DB records (not from user input); the API only accepts `node_exec_id` from users. Synthetic `copilot-*` IDs are only ever created server-side in `run_block.py`. The prefix skip avoids a DB lookup for a `AgentGraphExecution` record that legitimately does not exist for CoPilot sessions, while `user_id` scoping is enforced at the auth layer and on the resulting auto-approval record.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12873
File: autogpt_platform/backend/backend/copilot/baseline/reasoning.py:0-0
Timestamp: 2026-04-21T17:31:26.829Z
Learning: In `autogpt_platform/backend/backend/copilot/baseline/reasoning.py` (`BaselineReasoningEmitter`), when `render_in_ui=False`, BOTH the `StreamReasoning*` wire events AND the `ChatMessage(role="reasoning")` persistence append must be suppressed together. `convertChatSessionToUiMessages.ts` unconditionally re-renders all persisted `role="reasoning"` rows as `{type:"reasoning"}` UI parts on reload, so persisting rows while silencing live wire events would resurrect the reasoning collapse on page refresh. The audit trail is preserved through the provider transcript and `_format_sdk_content_blocks` (SDK path) instead. The baseline and SDK paths mirror each other: flag off → no live wire event, no persisted row, no hydrated collapse. This was established in PR `#12873`, commit 7ef10b26c.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12773
File: autogpt_platform/backend/backend/blocks/autopilot.py:631-638
Timestamp: 2026-04-14T07:35:11.464Z
Learning: In `autogpt_platform/backend/backend/copilot/executor/utils.py`, `CoPilotExecutionEntry` includes a `permissions: CopilotPermissions | None` field (added in PR `#12773` / commit a0184c87b9). `enqueue_copilot_turn` accepts and serializes this field into the queue entry, `_enqueue_for_recovery` in `autopilot.py` accepts and forwards `permissions` to `enqueue_copilot_turn`, and `_execute_async` in `processor.py` restores `entry.permissions` and passes it into `stream_chat_completion_sdk`/`stream_chat_completion_baseline` via `set_execution_context`. This ensures recovered sub-agent turns respect the same tool/block permission ceiling as the original in-process execution (mirroring `_merge_inherited_permissions`). Do NOT flag recovered turns as losing their permission ceiling — it is now fully propagated through the queue.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12814
File: autogpt_platform/backend/backend/copilot/model.py:0-0
Timestamp: 2026-04-16T13:28:28.641Z
Learning: In `autogpt_platform/backend/backend/copilot/model.py` (PR `#12814`, commit 259d37083): `append_and_save_message` uses `async with _get_session_lock(session_id)` — the same shared context manager used across the module — which internally acquires `redis-py`'s built-in `Lock` (key `copilot:session_lock:{session_id}`, timeout=10s, blocking_timeout=2s) via an atomic Lua-script. Lock release is also owner-verified via Lua so a slow pod can never delete a lock it no longer holds. On Redis failure the lock is skipped with a warning; the in-function idempotency check (`session.messages[-1].role` and `.content` comparison) still runs as a fallback. Do NOT expect a raw `redis.set(nx=True)` / `redis.delete()` pattern here — that intermediate approach was replaced in commit 259d37083.

📚 Learning: 2026-04-14T14:36:25.545Z

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12773
File: autogpt_platform/backend/backend/copilot/pending_messages.py:52-64
Timestamp: 2026-04-14T14:36:25.545Z
Learning: In `autogpt_platform/backend/backend/copilot` (PR `#12773`, commit d7bced0c6): when draining pending messages into `session.messages`, each message's text is sanitized via `strip_user_context_tags` before persistence to prevent user-controlled `<user_context>` injection from bypassing the trusted server-side context prefix. Additionally, if `upsert_chat_session` fails after draining, the drained `PendingMessage` objects are requeued back to Redis to avoid silent message loss. Do NOT flag the drain-then-requeue pattern as redundant — it is the intentional failure-resilience strategy for the pending buffer.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-03-11T08:40:59.673Z

Learnt from: kcze
Repo: Significant-Gravitas/AutoGPT PR: 12328
File: autogpt_platform/frontend/src/app/(platform)/copilot/useLoadMoreMessages.ts:49-61
Timestamp: 2026-03-11T08:40:59.673Z
Learning: In `autogpt_platform/frontend/src/app/(platform)/copilot/useLoadMoreMessages.ts`, clearing `olderMessages` (and resetting `oldestSequence`/`hasMore`) when `initialOldestSequence` shifts on the same session is intentional. Pages already fetched were based on a now-stale cursor; retaining them risks sequence gaps or duplicates. `ScrollPreserver` keeps the currently visible viewport intact, so only unvisited older pages are dropped. This is a deliberate safe-refetch design tradeoff.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-04-25T02:53:53.964Z

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 0
File: :0-0
Timestamp: 2026-04-25T02:53:53.964Z
Learning: In `autogpt_platform/backend/backend/copilot/sdk/service.py` (PR `#12918`, commit 6576bf561):
- `_flush_unresolved_tool_calls` was renamed to `flush_unresolved_tool_calls` (public); all call sites updated, `# noqa: SLF001` suppressor removed.
- `_flush_orphan_tool_uses_to_session` and `_InterruptedAttempt.finalize` both return `list[StreamBaseResponse]`; the post-loop caller yields those events directly to avoid double-flush and skipped UI cleanup events.
- The three former post-loop blocks (partial restore + redundant re-flush + two separate `yield StreamError` sites) are collapsed into a single block driven by `_classify_final_failure` returning a `_FinalFailure(display_msg, code, retryable)` dataclass, so history marker and SSE yield share one source of truth.
Do NOT flag double-flush risk or mismatched history/SSE marker as issues in the post-loop section of `stream_chat_completion_sdk`.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-04-30T03:25:37.606Z

Learnt from: CR
Repo: Significant-Gravitas/AutoGPT PR: 0
File: autogpt_platform/frontend/AGENTS.md:0-0
Timestamp: 2026-04-30T03:25:37.606Z
Learning: Applies to autogpt_platform/frontend/**/*.{ts,tsx} : Use function declarations (not arrow functions) for components/handlers

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-04-15T13:44:34.273Z

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12797
File: autogpt_platform/backend/backend/copilot/sdk/service.py:1991-2021
Timestamp: 2026-04-15T13:44:34.273Z
Learning: In `autogpt_platform/backend/backend/copilot/sdk/service.py` (`_run_stream_attempt`), the pre-create block (PR `#12797`) intentionally does NOT call `state.transcript_builder.append_assistant(...)` when inserting the empty assistant placeholder into `ctx.session.messages`. The transcript is left ending at the `tool_result` entry (N entries) while `message_count` metadata is N+1. This mismatch is benign and deliberate: on the next `--resume`, the SDK sees the transcript ending at `tool_result` and correctly regenerates the assistant response. Pre-appending the assistant turn to the transcript would suppress regeneration while leaving `session.messages[-1].content = ""` permanently (worse outcome). On the gap-fallback path, `transcript_msg_count (N+1) >= msg_count-1 (N)` means no gap is injected for the empty placeholder, which is correct because injecting an empty assistant message as context would mislead the SDK. Do NOT flag this transcript/message_count discrepancy as a bug.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-03-24T02:23:31.305Z

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12526
File: autogpt_platform/frontend/src/app/(platform)/copilot/components/RateLimitResetDialog/RateLimitResetDialog.tsx:0-0
Timestamp: 2026-03-24T02:23:31.305Z
Learning: In the Copilot platform UI code, follow the established Orval hook `onError` error-handling convention: first explicitly detect/handle `ApiError`, then read `error.response?.detail` (if present) as the primary message; if not available, fall back to `error.message`; and finally fall back to a generic string message. This convention should be used for generated Orval hooks even if the custom Orval mutator already maps details into `ApiError.message`, to keep consistency across hooks/components (e.g., `useCronSchedulerDialog.ts`, `useRunGraph.ts`, and rate-limit/reset flows).

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-04-01T18:54:16.035Z

Learnt from: Bentlybro
Repo: Significant-Gravitas/AutoGPT PR: 12633
File: autogpt_platform/frontend/src/app/(platform)/library/components/AgentFilterMenu/AgentFilterMenu.tsx:3-10
Timestamp: 2026-04-01T18:54:16.035Z
Learning: In the frontend, the legacy Select component at `@/components/__legacy__/ui/select` is an intentional, codebase-wide visual-consistency pattern. During code reviews, do not flag or block PRs merely for continuing to use this legacy Select. If a migration to the newer design-system Select is desired, bundle it into a single dedicated cleanup/migration PR that updates all Select usages together (e.g., avoid piecemeal replacements).

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-04-07T09:24:16.582Z

Learnt from: 0ubbe
Repo: Significant-Gravitas/AutoGPT PR: 12686
File: autogpt_platform/frontend/src/app/(no-navbar)/onboarding/steps/__tests__/PainPointsStep.test.tsx:1-19
Timestamp: 2026-04-07T09:24:16.582Z
Learning: In Significant-Gravitas/AutoGPT’s `autogpt_platform/frontend` (Vite + `vitejs/plugin-react` with the automatic JSX transform), do not flag usages of React types/components (e.g., `React.ReactNode`) in `.ts`/`.tsx` files as missing `React` imports. Since the React namespace is made available by the project’s TS/Vite setup, an explicit `import React from 'react'` or `import type { ReactNode } ...` is not required; only treat it as missing if typechecking (e.g., `pnpm types`) would actually fail.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-04-02T05:43:49.128Z

Learnt from: 0ubbe
Repo: Significant-Gravitas/AutoGPT PR: 12640
File: autogpt_platform/frontend/src/app/(no-navbar)/onboarding/steps/WelcomeStep.tsx:13-13
Timestamp: 2026-04-02T05:43:49.128Z
Learning: Do not flag `import { Question } from "phosphor-icons/react"` as an invalid import. `Question` is a valid named export from `phosphor-icons/react` (as reflected in the package’s generated `.d.ts` files and re-exports via `dist/index.d.ts`), so it should be treated as a supported named export during code reviews.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

[majdyz]

Addressed all three CodeRabbit findings in e59fe5af76:

1. TOCTOU + sanitised log in delete_stale_cli_session_file — direct unlink(), separate FileNotFoundError branch, log uses os.path.basename(real_path) + unlink_err.strerror only.
2. Unified bubble id between auto-continue and mid-turn promotion via new bubbleIdFor(chip) = pending-chip-${chip.id} helper — both paths now produce the same id for the same chip, so the dedup filter actually catches race-promoted dupes.
3. Stale-peek session-switch guard — capture requestSessionId at request time and verify prevSessionIdRef.current === requestSessionId before mutating chip state in the .then() callback.

Tests still green (216 backend SDK + 29 focused frontend) and pre-commit checks (black/ruff/pnpm format/lint/types) all clean.

[coderabbitai]

Actionable comments posted: 1

🧹 Nitpick comments (1)

autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts (1)

3-3: ⚡ Quick win

Remove useMemo/useCallback here to match frontend hook conventions.

This introduces optimization hooks where the repo guideline says to avoid them unless explicitly requested.

Suggested fix

-import { useCallback, useEffect, useMemo, useRef, useState } from "react";
+import { useEffect, useRef, useState } from "react";
@@
-  const queuedMessages = useMemo(() => chips.map((c) => c.text), [chips]);
+  const queuedMessages = chips.map((c) => c.text);
@@
-  const appendChip = useCallback((text: string) => {
+  function appendChip(text: string) {
     setChips((prev) => [...prev, { id: crypto.randomUUID(), text }]);
-  }, []);
+  }

As per coding guidelines, “Do not use useCallback or useMemo unless asked to optimize a given function.”

Also applies to: 58-61, 81-83

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In
`@autogpt_platform/frontend/src/app/`(platform)/copilot/useCopilotPendingChips.ts
at line 3, The file's hook imports and usages of useMemo/useCallback should be
removed to follow the repo convention: drop useMemo and useCallback from the
import list and replace any memoized values and callbacks inside
useCopilotPendingChips with plain functions/values (e.g., convert memoized
selectors and callbacks at the locations noted—previously using
useCallback/useMemo around the logic at the ~58-61 and ~81-83 areas—into regular
functions or computed values inside the hook), ensuring you adjust any
references and remove unnecessary dependency arrays; keep
useRef/useState/useEffect as needed.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Inline comments:
In
`@autogpt_platform/frontend/src/app/`(platform)/copilot/useCopilotPendingChips.ts:
- Around line 284-286: pollBackendAndPromote may apply results from an old
session after an async wait; guard against session switches by checking that the
active session still matches before calling state setters. Specifically, after
awaiting pollBackendAndPromote (or inside pollBackendAndPromote before calling
setMessages/setChips), compare the original sessionId param with the current
session identifier (e.g., from the hook or getCurrentSession function) and bail
out if they differ so promoted chips/messages are not applied to a new session;
update the interval callback and any other async paths in useCopilotPendingChips
(including the block spanning the pollBackendAndPromote usage around lines
291–344) to perform this post-await session check.

---

Nitpick comments:
In
`@autogpt_platform/frontend/src/app/`(platform)/copilot/useCopilotPendingChips.ts:
- Line 3: The file's hook imports and usages of useMemo/useCallback should be
removed to follow the repo convention: drop useMemo and useCallback from the
import list and replace any memoized values and callbacks inside
useCopilotPendingChips with plain functions/values (e.g., convert memoized
selectors and callbacks at the locations noted—previously using
useCallback/useMemo around the logic at the ~58-61 and ~81-83 areas—into regular
functions or computed values inside the hook), ensuring you adjust any
references and remove unnecessary dependency arrays; keep
useRef/useState/useEffect as needed.

🪄 Autofix (Beta)

Fix all unresolved CodeRabbit comments on this PR:

• Push a commit to this branch (recommended)
• Create a new PR with the fixes

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 4292532d-b157-4a05-ab85-0294ec755ae5

📥 Commits

Reviewing files that changed from the base of the PR and between 9e8622c and e59fe5a.

📒 Files selected for processing (2)

• autogpt_platform/backend/backend/copilot/sdk/service.py
• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

🚧 Files skipped from review as they are similar to previous changes (1)

• autogpt_platform/backend/backend/copilot/sdk/service.py

📜 Review details

⏰ Context from checks skipped due to timeout of 90000ms. You can increase the timeout in your CodeRabbit configuration to a maximum of 15 minutes (900000ms). (13)

• GitHub Check: integration_test
• GitHub Check: lint
• GitHub Check: check API types
• GitHub Check: type-check (3.11)
• GitHub Check: type-check (3.12)
• GitHub Check: test (3.12)
• GitHub Check: type-check (3.13)
• GitHub Check: test (3.13)
• GitHub Check: test (3.11)
• GitHub Check: Analyze (python)
• GitHub Check: end-to-end tests
• GitHub Check: Analyze (typescript)
• GitHub Check: Check PR Status

🧰 Additional context used

📓 Path-based instructions (8)

autogpt_platform/frontend/**/*.{ts,tsx,js,jsx}

📄 CodeRabbit inference engine (.github/copilot-instructions.md)

autogpt_platform/frontend/**/*.{ts,tsx,js,jsx}: Use Node.js 21+ with pnpm package manager for frontend development
Always run 'pnpm format' for formatting and linting code in frontend development

Format frontend code using pnpm format

autogpt_platform/frontend/**/*.{ts,tsx,js,jsx}: Fully capitalize acronyms in symbols, e.g. graphID, useBackendAPI
No linter suppressors (// @ts-ignore``, // eslint-disable) — fix the actual issue

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

autogpt_platform/frontend/**/*.{tsx,ts}

📄 CodeRabbit inference engine (.github/copilot-instructions.md)

autogpt_platform/frontend/**/*.{tsx,ts}: Use function declarations for components and handlers (not arrow functions) in React components
Only use arrow functions for small inline lambdas (map, filter, etc.) in React components
Use PascalCase for component names and camelCase with 'use' prefix for hook names in React
Use Tailwind CSS utilities only for styling in frontend components
Use design system components from 'src/components/' (atoms, molecules, organisms) in frontend development
Never use 'src/components/legacy/' in frontend code
Only use Phosphor Icons (@phosphor-icons/react) for icons in frontend components
Use generated API hooks from '@/app/api/generated/endpoints/' instead of deprecated 'BackendAPI' or 'src/lib/autogpt-server-api/'
Use React Query for server state (via generated hooks) in frontend development
Default to client components ('use client') in Next.js; only use server components for SEO or extreme TTFB needs
Use '' component for rendering errors in frontend UI; use toast notifications for mutation errors; use 'Sentry.captureException()' for manual exceptions
Separate render logic from data/behavior in React components; keep comments minimal (code should be self-documenting)

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

autogpt_platform/frontend/**/*.{ts,tsx}

📄 CodeRabbit inference engine (.github/copilot-instructions.md)

autogpt_platform/frontend/**/*.{ts,tsx}: No barrel files or 'index.ts' re-exports in frontend code
Regenerate API hooks with 'pnpm generate:api' after backend OpenAPI spec changes in frontend development

autogpt_platform/frontend/**/*.{ts,tsx}: Use function declarations (not arrow functions) for components/handlers
No any types unless the value genuinely can be anything
Keep render functions and hooks under ~50 lines; extract named helpers or sub-components when they grow longer

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

autogpt_platform/frontend/src/**/*.{ts,tsx}

📄 CodeRabbit inference engine (AGENTS.md)

autogpt_platform/frontend/src/**/*.{ts,tsx}: Use generated API hooks from @/app/api/__generated__/endpoints/ following the pattern use{Method}{Version}{OperationName}, and regenerate with pnpm generate:api
Separate render logic from business logic using component.tsx + useComponent.ts + helpers.ts pattern, colocate state when possible and avoid creating large components, use sub-components in local /components folder
Use function declarations for components and handlers, use arrow functions only for callbacks
Do not use useCallback or useMemo unless asked to optimise a given function

autogpt_platform/frontend/src/**/*.{ts,tsx}: Keep files under ~200 lines; extract sub-components or hooks into their own files when a file grows beyond this
Use generated API hooks from @/app/api/__generated__/endpoints/ with pattern use{Method}{Version}{OperationName}
Always import the -Icon-suffixed alias from @phosphor-icons/react (e.g. TrashIcon, PlusIcon, SquareIcon) — bare exports are deprecated
Do not use useCallback or useMemo unless asked to optimize a given function
Never use src/components/__legacy__/* — use design system components from src/components/

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

autogpt_platform/frontend/**/*.ts

📄 CodeRabbit inference engine (AGENTS.md)

No barrel files or index.ts re-exports in the frontend

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

autogpt_platform/frontend/src/**/*.ts

📄 CodeRabbit inference engine (AGENTS.md)

Do not type hook returns, let Typescript infer as much as possible

autogpt_platform/frontend/src/**/*.ts: Extract component logic into custom hooks grouped by concern, not by component, with each hook in its own .ts file
Do not type hook returns; let TypeScript infer as much as possible

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

autogpt_platform/**/*.{ts,tsx}

📄 CodeRabbit inference engine (AGENTS.md)

Never type with any, if no types available use unknown

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

autogpt_platform/frontend/src/**/*.{ts,tsx,js,jsx}

📄 CodeRabbit inference engine (autogpt_platform/frontend/AGENTS.md)

Avoid index and barrel files

Files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

🧠 Learnings (14)

📓 Common learnings

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12773
File: autogpt_platform/backend/backend/copilot/pending_messages.py:52-64
Timestamp: 2026-04-14T14:36:25.545Z
Learning: In `autogpt_platform/backend/backend/copilot` (PR `#12773`, commit d7bced0c6): when draining pending messages into `session.messages`, each message's text is sanitized via `strip_user_context_tags` before persistence to prevent user-controlled `<user_context>` injection from bypassing the trusted server-side context prefix. Additionally, if `upsert_chat_session` fails after draining, the drained `PendingMessage` objects are requeued back to Redis to avoid silent message loss. Do NOT flag the drain-then-requeue pattern as redundant — it is the intentional failure-resilience strategy for the pending buffer.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 0
File: :0-0
Timestamp: 2026-04-25T02:53:53.964Z
Learning: In `autogpt_platform/backend/backend/copilot/sdk/service.py` (PR `#12918`, commit 6576bf561):
- `_flush_unresolved_tool_calls` was renamed to `flush_unresolved_tool_calls` (public); all call sites updated, `# noqa: SLF001` suppressor removed.
- `_flush_orphan_tool_uses_to_session` and `_InterruptedAttempt.finalize` both return `list[StreamBaseResponse]`; the post-loop caller yields those events directly to avoid double-flush and skipped UI cleanup events.
- The three former post-loop blocks (partial restore + redundant re-flush + two separate `yield StreamError` sites) are collapsed into a single block driven by `_classify_final_failure` returning a `_FinalFailure(display_msg, code, retryable)` dataclass, so history marker and SSE yield share one source of truth.
Do NOT flag double-flush risk or mismatched history/SSE marker as issues in the post-loop section of `stream_chat_completion_sdk`.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12445
File: autogpt_platform/backend/backend/copilot/sdk/service.py:1071-1072
Timestamp: 2026-03-17T06:48:26.471Z
Learning: In Significant-Gravitas/AutoGPT (autogpt_platform), the AI SDK enforces `z.strictObject({type, errorText})` on SSE `StreamError` responses, so additional fields like `retryable: bool` cannot be added to `StreamError` or serialized via `to_sse()`. Instead, retry signaling for transient Anthropic API errors is done via the `COPILOT_RETRYABLE_ERROR_PREFIX` constant prepended to persisted session messages (in `ChatMessage.content`). The frontend detects retryable errors by checking `markerType === "retryable_error"` from `parseSpecialMarkers()` — no SSE schema changes and no string matching on error text. This pattern was established in PR `#12445`, commit 64d82797b.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12796
File: autogpt_platform/backend/backend/api/features/chat/routes.py:504-527
Timestamp: 2026-04-16T12:33:44.990Z
Learning: In `autogpt_platform/backend/backend/api/features/chat/routes.py`, `get_session` (PR `#12796`, commit 3771bfad9c1) closes the TOCTOU race between the initial `stream_registry.get_active_session()` pre-check and `get_chat_messages_paginated()` with a post-check re-verification: after the DB fetch, if `is_initial_load and active_session is not None`, it calls `get_active_session` a second time; if `post_active is None` (stream completed during the window), it resets `from_start=True`, `forward_paginated=True`, and re-fetches messages from sequence 0. Do NOT flag the double `get_active_session` call pattern as redundant — it is the intentional TOCTOU mitigation for pagination direction selection.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12797
File: autogpt_platform/backend/backend/copilot/sdk/service.py:1991-2021
Timestamp: 2026-04-15T13:44:34.273Z
Learning: In `autogpt_platform/backend/backend/copilot/sdk/service.py` (`_run_stream_attempt`), the pre-create block (PR `#12797`) intentionally does NOT call `state.transcript_builder.append_assistant(...)` when inserting the empty assistant placeholder into `ctx.session.messages`. The transcript is left ending at the `tool_result` entry (N entries) while `message_count` metadata is N+1. This mismatch is benign and deliberate: on the next `--resume`, the SDK sees the transcript ending at `tool_result` and correctly regenerates the assistant response. Pre-appending the assistant turn to the transcript would suppress regeneration while leaving `session.messages[-1].content = ""` permanently (worse outcome). On the gap-fallback path, `transcript_msg_count (N+1) >= msg_count-1 (N)` means no gap is injected for the empty placeholder, which is correct because injecting an empty assistant message as context would mislead the SDK. Do NOT flag this transcript/message_count discrepancy as a bug.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12814
File: autogpt_platform/backend/backend/copilot/model.py:0-0
Timestamp: 2026-04-16T13:28:28.641Z
Learning: In `autogpt_platform/backend/backend/copilot/model.py` (PR `#12814`, commit 259d37083): `append_and_save_message` uses `async with _get_session_lock(session_id)` — the same shared context manager used across the module — which internally acquires `redis-py`'s built-in `Lock` (key `copilot:session_lock:{session_id}`, timeout=10s, blocking_timeout=2s) via an atomic Lua-script. Lock release is also owner-verified via Lua so a slow pod can never delete a lock it no longer holds. On Redis failure the lock is skipped with a warning; the in-function idempotency check (`session.messages[-1].role` and `.content` comparison) still runs as a fallback. Do NOT expect a raw `redis.set(nx=True)` / `redis.delete()` pattern here — that intermediate approach was replaced in commit 259d37083.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12356
File: autogpt_platform/backend/backend/copilot/constants.py:9-12
Timestamp: 2026-03-10T08:39:22.025Z
Learning: In Significant-Gravitas/AutoGPT PR `#12356`, the `COPILOT_SYNTHETIC_ID_PREFIX = "copilot-"` check in `create_auto_approval_record` (human_review.py) is intentional and safe. The `graph_exec_id` passed to this function comes from server-side `PendingHumanReview` DB records (not from user input); the API only accepts `node_exec_id` from users. Synthetic `copilot-*` IDs are only ever created server-side in `run_block.py`. The prefix skip avoids a DB lookup for a `AgentGraphExecution` record that legitimately does not exist for CoPilot sessions, while `user_id` scoping is enforced at the auth layer and on the resulting auto-approval record.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 0
File: :0-0
Timestamp: 2026-04-28T03:31:29.696Z
Learning: In Significant-Gravitas/AutoGPT PR `#12933` (`fix/stripe-checkout-link-auth-loop`), the initial approach of pinning `payment_method_types=["card"]` in `top_up_intent` and `create_subscription_checkout` (in `autogpt_platform/backend/backend/data/credit.py`) was reverted in commit `584b43a71` as it patched a symptom. The true root cause was in `update_subscription_tier()` in `v1.py`: a `current_tier_price_id is not None` guard was gating admin-granted DB-tier flips and short-circuiting them when the BUSINESS tier was pruned from the price-id LaunchDarkly flag. Do NOT flag `payment_method_types` absence in these checkout helpers as a Stripe Link bypass issue; the fix lives in the subscription tier update guard logic.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12773
File: autogpt_platform/backend/backend/blocks/autopilot.py:631-638
Timestamp: 2026-04-14T07:35:11.464Z
Learning: In `autogpt_platform/backend/backend/copilot/executor/utils.py`, `CoPilotExecutionEntry` includes a `permissions: CopilotPermissions | None` field (added in PR `#12773` / commit a0184c87b9). `enqueue_copilot_turn` accepts and serializes this field into the queue entry, `_enqueue_for_recovery` in `autopilot.py` accepts and forwards `permissions` to `enqueue_copilot_turn`, and `_execute_async` in `processor.py` restores `entry.permissions` and passes it into `stream_chat_completion_sdk`/`stream_chat_completion_baseline` via `set_execution_context`. This ensures recovered sub-agent turns respect the same tool/block permission ceiling as the original in-process execution (mirroring `_merge_inherited_permissions`). Do NOT flag recovered turns as losing their permission ceiling — it is now fully propagated through the queue.

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12873
File: autogpt_platform/backend/backend/copilot/baseline/reasoning.py:0-0
Timestamp: 2026-04-21T17:31:26.829Z
Learning: In `autogpt_platform/backend/backend/copilot/baseline/reasoning.py` (`BaselineReasoningEmitter`), when `render_in_ui=False`, BOTH the `StreamReasoning*` wire events AND the `ChatMessage(role="reasoning")` persistence append must be suppressed together. `convertChatSessionToUiMessages.ts` unconditionally re-renders all persisted `role="reasoning"` rows as `{type:"reasoning"}` UI parts on reload, so persisting rows while silencing live wire events would resurrect the reasoning collapse on page refresh. The audit trail is preserved through the provider transcript and `_format_sdk_content_blocks` (SDK path) instead. The baseline and SDK paths mirror each other: flag off → no live wire event, no persisted row, no hydrated collapse. This was established in PR `#12873`, commit 7ef10b26c.

📚 Learning: 2026-04-14T14:36:25.545Z

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12773
File: autogpt_platform/backend/backend/copilot/pending_messages.py:52-64
Timestamp: 2026-04-14T14:36:25.545Z
Learning: In `autogpt_platform/backend/backend/copilot` (PR `#12773`, commit d7bced0c6): when draining pending messages into `session.messages`, each message's text is sanitized via `strip_user_context_tags` before persistence to prevent user-controlled `<user_context>` injection from bypassing the trusted server-side context prefix. Additionally, if `upsert_chat_session` fails after draining, the drained `PendingMessage` objects are requeued back to Redis to avoid silent message loss. Do NOT flag the drain-then-requeue pattern as redundant — it is the intentional failure-resilience strategy for the pending buffer.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-03-11T08:40:59.673Z

Learnt from: kcze
Repo: Significant-Gravitas/AutoGPT PR: 12328
File: autogpt_platform/frontend/src/app/(platform)/copilot/useLoadMoreMessages.ts:49-61
Timestamp: 2026-03-11T08:40:59.673Z
Learning: In `autogpt_platform/frontend/src/app/(platform)/copilot/useLoadMoreMessages.ts`, clearing `olderMessages` (and resetting `oldestSequence`/`hasMore`) when `initialOldestSequence` shifts on the same session is intentional. Pages already fetched were based on a now-stale cursor; retaining them risks sequence gaps or duplicates. `ScrollPreserver` keeps the currently visible viewport intact, so only unvisited older pages are dropped. This is a deliberate safe-refetch design tradeoff.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-04-25T02:53:53.964Z

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 0
File: :0-0
Timestamp: 2026-04-25T02:53:53.964Z
Learning: In `autogpt_platform/backend/backend/copilot/sdk/service.py` (PR `#12918`, commit 6576bf561):
- `_flush_unresolved_tool_calls` was renamed to `flush_unresolved_tool_calls` (public); all call sites updated, `# noqa: SLF001` suppressor removed.
- `_flush_orphan_tool_uses_to_session` and `_InterruptedAttempt.finalize` both return `list[StreamBaseResponse]`; the post-loop caller yields those events directly to avoid double-flush and skipped UI cleanup events.
- The three former post-loop blocks (partial restore + redundant re-flush + two separate `yield StreamError` sites) are collapsed into a single block driven by `_classify_final_failure` returning a `_FinalFailure(display_msg, code, retryable)` dataclass, so history marker and SSE yield share one source of truth.
Do NOT flag double-flush risk or mismatched history/SSE marker as issues in the post-loop section of `stream_chat_completion_sdk`.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-04-21T17:31:26.829Z

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12873
File: autogpt_platform/backend/backend/copilot/baseline/reasoning.py:0-0
Timestamp: 2026-04-21T17:31:26.829Z
Learning: In `autogpt_platform/backend/backend/copilot/baseline/reasoning.py` (`BaselineReasoningEmitter`), when `render_in_ui=False`, BOTH the `StreamReasoning*` wire events AND the `ChatMessage(role="reasoning")` persistence append must be suppressed together. `convertChatSessionToUiMessages.ts` unconditionally re-renders all persisted `role="reasoning"` rows as `{type:"reasoning"}` UI parts on reload, so persisting rows while silencing live wire events would resurrect the reasoning collapse on page refresh. The audit trail is preserved through the provider transcript and `_format_sdk_content_blocks` (SDK path) instead. The baseline and SDK paths mirror each other: flag off → no live wire event, no persisted row, no hydrated collapse. This was established in PR `#12873`, commit 7ef10b26c.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-03-17T06:48:26.471Z

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12445
File: autogpt_platform/backend/backend/copilot/sdk/service.py:1071-1072
Timestamp: 2026-03-17T06:48:26.471Z
Learning: In Significant-Gravitas/AutoGPT (autogpt_platform), the AI SDK enforces `z.strictObject({type, errorText})` on SSE `StreamError` responses, so additional fields like `retryable: bool` cannot be added to `StreamError` or serialized via `to_sse()`. Instead, retry signaling for transient Anthropic API errors is done via the `COPILOT_RETRYABLE_ERROR_PREFIX` constant prepended to persisted session messages (in `ChatMessage.content`). The frontend detects retryable errors by checking `markerType === "retryable_error"` from `parseSpecialMarkers()` — no SSE schema changes and no string matching on error text. This pattern was established in PR `#12445`, commit 64d82797b.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-03-17T06:18:51.570Z

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12445
File: autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatContainer/ChatContainer.tsx:55-67
Timestamp: 2026-03-17T06:18:51.570Z
Learning: In `autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatContainer/ChatContainer.tsx`, an explicit `isBusy` guard on the retry handler (`handleRetry`) is not needed. Once `onSend` is invoked, the chat status immediately transitions to "submitted", which causes the `ErrorCard` (containing the retry button) to unmount before a second click can register, making double-send impossible by design.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-04-13T13:11:00.401Z

Learnt from: 0ubbe
Repo: Significant-Gravitas/AutoGPT PR: 12764
File: autogpt_platform/frontend/src/app/(platform)/copilot/components/EmptySession/EmptySession.tsx:41-42
Timestamp: 2026-04-13T13:11:00.401Z
Learning: In Significant-Gravitas/AutoGPT `autogpt_platform/frontend`, unconditional React Query hook calls (e.g. `usePulseChips()` in `EmptySession.tsx`) are intentional when the underlying data is expected to be cached from prior page visits. The team considers the fetch cost acceptable in these cases and does not require `enabled` gating purely for feature-flag-disabled paths. Do not flag unconditional query hooks as wasteful when caching makes the cost negligible.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-04-15T13:44:34.273Z

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12797
File: autogpt_platform/backend/backend/copilot/sdk/service.py:1991-2021
Timestamp: 2026-04-15T13:44:34.273Z
Learning: In `autogpt_platform/backend/backend/copilot/sdk/service.py` (`_run_stream_attempt`), the pre-create block (PR `#12797`) intentionally does NOT call `state.transcript_builder.append_assistant(...)` when inserting the empty assistant placeholder into `ctx.session.messages`. The transcript is left ending at the `tool_result` entry (N entries) while `message_count` metadata is N+1. This mismatch is benign and deliberate: on the next `--resume`, the SDK sees the transcript ending at `tool_result` and correctly regenerates the assistant response. Pre-appending the assistant turn to the transcript would suppress regeneration while leaving `session.messages[-1].content = ""` permanently (worse outcome). On the gap-fallback path, `transcript_msg_count (N+1) >= msg_count-1 (N)` means no gap is injected for the empty placeholder, which is correct because injecting an empty assistant message as context would mislead the SDK. Do NOT flag this transcript/message_count discrepancy as a bug.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-03-10T08:39:22.025Z

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12356
File: autogpt_platform/backend/backend/copilot/constants.py:9-12
Timestamp: 2026-03-10T08:39:22.025Z
Learning: In Significant-Gravitas/AutoGPT PR `#12356`, the `COPILOT_SYNTHETIC_ID_PREFIX = "copilot-"` check in `create_auto_approval_record` (human_review.py) is intentional and safe. The `graph_exec_id` passed to this function comes from server-side `PendingHumanReview` DB records (not from user input); the API only accepts `node_exec_id` from users. Synthetic `copilot-*` IDs are only ever created server-side in `run_block.py`. The prefix skip avoids a DB lookup for a `AgentGraphExecution` record that legitimately does not exist for CoPilot sessions, while `user_id` scoping is enforced at the auth layer and on the resulting auto-approval record.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-03-24T02:23:31.305Z

Learnt from: majdyz
Repo: Significant-Gravitas/AutoGPT PR: 12526
File: autogpt_platform/frontend/src/app/(platform)/copilot/components/RateLimitResetDialog/RateLimitResetDialog.tsx:0-0
Timestamp: 2026-03-24T02:23:31.305Z
Learning: In the Copilot platform UI code, follow the established Orval hook `onError` error-handling convention: first explicitly detect/handle `ApiError`, then read `error.response?.detail` (if present) as the primary message; if not available, fall back to `error.message`; and finally fall back to a generic string message. This convention should be used for generated Orval hooks even if the custom Orval mutator already maps details into `ApiError.message`, to keep consistency across hooks/components (e.g., `useCronSchedulerDialog.ts`, `useRunGraph.ts`, and rate-limit/reset flows).

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-04-01T18:54:16.035Z

Learnt from: Bentlybro
Repo: Significant-Gravitas/AutoGPT PR: 12633
File: autogpt_platform/frontend/src/app/(platform)/library/components/AgentFilterMenu/AgentFilterMenu.tsx:3-10
Timestamp: 2026-04-01T18:54:16.035Z
Learning: In the frontend, the legacy Select component at `@/components/__legacy__/ui/select` is an intentional, codebase-wide visual-consistency pattern. During code reviews, do not flag or block PRs merely for continuing to use this legacy Select. If a migration to the newer design-system Select is desired, bundle it into a single dedicated cleanup/migration PR that updates all Select usages together (e.g., avoid piecemeal replacements).

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-04-07T09:24:16.582Z

Learnt from: 0ubbe
Repo: Significant-Gravitas/AutoGPT PR: 12686
File: autogpt_platform/frontend/src/app/(no-navbar)/onboarding/steps/__tests__/PainPointsStep.test.tsx:1-19
Timestamp: 2026-04-07T09:24:16.582Z
Learning: In Significant-Gravitas/AutoGPT’s `autogpt_platform/frontend` (Vite + `vitejs/plugin-react` with the automatic JSX transform), do not flag usages of React types/components (e.g., `React.ReactNode`) in `.ts`/`.tsx` files as missing `React` imports. Since the React namespace is made available by the project’s TS/Vite setup, an explicit `import React from 'react'` or `import type { ReactNode } ...` is not required; only treat it as missing if typechecking (e.g., `pnpm types`) would actually fail.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

📚 Learning: 2026-04-02T05:43:49.128Z

Learnt from: 0ubbe
Repo: Significant-Gravitas/AutoGPT PR: 12640
File: autogpt_platform/frontend/src/app/(no-navbar)/onboarding/steps/WelcomeStep.tsx:13-13
Timestamp: 2026-04-02T05:43:49.128Z
Learning: Do not flag `import { Question } from "phosphor-icons/react"` as an invalid import. `Question` is a valid named export from `phosphor-icons/react` (as reflected in the package’s generated `.d.ts` files and re-exports via `dist/index.d.ts`), so it should be treated as a supported named export during code reviews.

Applied to files:

• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts

🔇 Additional comments (1)

autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPendingChips.ts (1)

228-234: Deterministic chip bubble IDs and dedup filtering look solid.

Using pending-chip-${chip.id} across both promotion paths cleanly prevents duplicate renders when effects re-run or poll ordering varies.

Also applies to: 244-251, 328-334

[majdyz]

/pr-test --fix result

TL;DR

PASS on unit/integration coverage, UI smoke test BLOCKED by an unrelated native-backend startup bug (TypeError: cannot pickle '_thread.lock' object from backend/app.py:22 → multiprocessing/popen_forkserver.py). Two background test agents stalled on the same bring-up. Lock released.

What was actually verified

Scenario	Coverage	Trigger
1. Disappearing queued messages	✅ unit	Race-safe path: useCopilotPendingChips.test.ts adds a synthetic test "mid-turn poll: chip appended during in-flight poll survives the drain" — uses vi.useFakeTimers() + a deferred peek promise (new Promise(resolve => resolvePeek = resolve)) to hold the GET in-flight, then appendChip("chipB") while it's pending, then resolves the peek with count: 0 and asserts only chipA was promoted and queuedMessages === ["chipB"].
2. Prompt-too-long persistence	✅ unit + integration	service_helpers_test.py::TestSdkSessionIdSelection::test_retry_keeps_session_id_for_t2_plus asserts the retry kwargs now keep session_id (regression for the SENTRY-1207 cause). TestDeleteStaleCliSessionFile covers the helper end-to-end via tmp_path — patches cli_session_path + projects_base and asserts: file deleted when present, returns False when missing, returns False on path-traversal violation.
3. Double error UI	✅ unit	ChatMessagesContainer.test.tsx adds a "error banner dedup" describe block: renders the component with error={new Error(...)} plus a persisted assistant message containing [__COPILOT_ERROR_f7a1__] and asserts screen.queryByText(/encountered an error/i) returns null. Plus the same for the retryable variant, plus the inverse (no marker → banner shown).
All targeted backend SDK tests	✅	poetry run pytest backend/copilot/sdk/{service_helpers,retry_scenarios,prompt_too_long,session_persistence,context_fallback}_test.py → 216 passed
All focused frontend tests	✅	pnpm vitest run on the three changed files → 29 passed
lint / format / types	✅	pnpm format/lint/types, poetry run black + ruff check — clean

How the harder scenarios would be triggered synthetically (UI not run)

These weren't executable today because the native backend won't start. If you re-run when the multiprocess bug is fixed:

Scenario 1 (disappearing chips, live): open the copilot, send a message that triggers a slow tool call (e.g. find_block on a big workspace) so the assistant streams for >10s. Click the queue (↑) button twice in quick succession with different texts. Observe that both chips remain visible until promoted. Race window can be widened by temporarily setting MID_TURN_POLL_MS = 200 in useCopilotPendingChips.ts.

Scenario 2 (prompt-too-long, live): simplest synthetic trigger is CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=20 in the backend .env (drops the autocompact threshold so a normal-length conversation hits it after ~5–10 turns). Run a session, observe the recovery turn succeeds, observe the local ~/.claude/projects/<encoded-cwd>/<session_id>.jsonl file shrinks (i.e. the recovery's compacted content was written). Force-reload the page; next turn should NOT re-trip prompt-too-long. Pre-fix: it would re-trip; post-fix: it doesn't.

Alternative: pre-populate session.messages in Postgres via the chat_db accessor with a 100k-token transcript, then start a new turn — first attempt errors, retry compacts and succeeds, GCS now holds the compacted JSONL.

Scenario 3 (double error UI, live): force a transient backend error (kill the backend mid-stream, or block the upstream Anthropic API at the firewall). Observe the in-line ErrorCard bubble appears and the trailing red banner does NOT (was: both appeared with identical text).

What blocked the live UI run

File "/Users/majdyz/Code/AutoGPT13/autogpt_platform/backend/backend/app.py", line 22, in run_processes
    process.start(background=True, **kwargs)
  ...
  File "/.../multiprocessing/popen_forkserver.py", line 47, in _launch
    reduction.dump(process_obj, buf)
TypeError: cannot pickle '_thread.lock' object

The native poetry run app process tree fails to fork — one of the manager-style services holds a non-pickleable lock when forkserver tries to pass process_obj. Unrelated to anything in this PR; pre-existing on dev. Both /pr-test --fix background agents stalled here too. Lock released; if you want a live UI verification, drop me a note once the native stack starts again.

[majdyz]

/pr-test --fix result (native stack)

Both UI smoke scenarios PASSED end-to-end after diagnosing + fixing the unrelated native-bring-up blocker (kept on this PR — under 20% scope, same module). Stack still running on :3000 / :8006.

Native bring-up

Step	Status
Pickle blocker (was: cannot pickle '_thread.lock' object)	RESOLVED in 9969b1ac07
Backend :8006 (poetry run app, all 9 subprocesses)	UP
Frontend :3000 (pnpm dev)	UP
Auth via local supabase	OK

Pickle blocker root-cause + fix

backend/copilot/executor/manager.py:88 instantiated self._active_tasks_lock = threading.Lock() eagerly in __init__ (added in #12877 to make _cleanup_completed_tasks thread-safe). With forkserver/spawn start method, Process(target=self.execute_run_command).start() (backend/util/process.py:133) tries to pickle the AppProcess instance — and _thread.lock is unpicklable. Result: TypeError: cannot pickle '_thread.lock' object aborts startup before REST server binds.

Fix: convert to a lazy @property _active_tasks_lock (mirrors the existing pattern already used for executor, *_thread, *_client, stop_consuming in the same class). Lock is materialized inside the forked child on first use. 32/32 executor tests still pass.

Smoke tests

#	Scenario	Result	Evidence
1	Disappearing chips (THE headline)	PASS	Both first follow up question + second follow up question chips persisted as separate "Queued" rows mid-stream, then drained into separate user bubbles after stream completion. The {id, text}[] + functional-updater drain pattern works.
2	Double error UI (huge prompt to force Prompt is too long)	PASS	Single error UI element on page (verified via DOM query for both [role=alert] and any "encountered an error"/"Prompt is too long"/"too long" text → {"err_count":1, "error_text_matches":1}). The lastAssistantHasErrorMarker gate works — no twin red-banner-plus-bubble.
3	Prompt-too-long persistence (cross-turn)	SKIPPED	Not realistically reproducible in a single session (would need ~50+ turns or pre-populated DB transcript). Unit test coverage at TestSdkSessionIdSelection::test_retry_keeps_session_id_for_t2_plus + TestDeleteStaleCliSessionFile is the regression guard.

Synthetic-trigger plan (for future re-verification of scenario 3)

When you want to live-validate the prompt-too-long persistence fix:

• Synthetic option A: set CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=20 in backend/.env so the autocompact threshold drops; ~5–10 turns naturally exceeds it. Watch ~/.claude/projects/<encoded-cwd>/<session_id>.jsonl shrink across the recovery turn (proves the rescued content was written to the predictable path) → next turn doesn't re-trip.
• Synthetic option B: pre-populate session.messages in Postgres via the chat_db accessor with a 100k-token transcript, then start a new turn — first attempt errors, retry compacts and succeeds, GCS now holds the compacted JSONL.

Screenshots

/Users/majdyz/Code/AutoGPT13/test-results/PR-12948-fix-copilot-stream-errors-and-queue-bubbles/:

• 06-both-chips-queued.png — both chips visible side-by-side, both labelled "Queued" (headline scenario PASS)
• 07-after-30s.png — both chips drained into separate user bubbles
• 09-after-huge-submit.png — single trailing error banner, no duplicate

Lock state

Released — /Users/majdyz/Code/AutoGPT/.ign.testing.lock removed; release line appended to .ign.testing.log. Native stack left running.

[majdyz]

Live UI verification - PR #12948 dedup fix on dev preview

• HEAD tested: 93aa3dd25a (per request) — dev k8s actually serves b35a49a5b9 which is 93aa3dd + 83b5765b22 (transport-side messageId fix) + b35a49a5b9 (transport regression test). All 3 commits are on this PR's branch.
• Frontend host used: https://dev-builder.agpt.co — the Vercel preview at autogpt-11dxfg7pp-significant-gravitas.vercel.app would not authenticate API calls (cross-origin auth tokens not propagating from vercel.app → dev-server.agpt.co; UI rendered but every /api/proxy/... returned without an Authorization header). Both hosts deploy the same PR build, so the fix logic is the same; only the frontend container differs.
• Backend (dev k8s): confirmed serving the PR branch by Sentry release 93aa3dd25a2f4d5e732a4b052504fef1071017ad on the page and by git show 93aa3dd25a -- routes.py matching the deployed behavior.

Per-scenario results

#	Scenario	Result	Evidence
1	Baseline send → assistant streams PR12948-DEVPREVIEW-OK	PASS	Reply rendered verbatim.
2	message_id UUID is plumbed into the stream POST body	PASS	Captured POST /api/chat/sessions/2635be7a.../stream body contains "message_id":"ed2ca707-de7d-4c9c-9f57-e80dcdecedf7" (UUIDv4).
3	Concurrent identical POST → PK collision short-circuit	PASS	Fired two concurrent fetch() with the same message_id=956e11c0-7a10-4fd5-ace6-4e0279e9b12f. Both returned 200, but both responses were empty SSE streams (start → finish → [DONE] with no model output) and 956e11c0-... does not appear in GET /sessions/{id} user-row list — exactly the dedup behavior. Backend log excerpt: [STREAM] Duplicate message detected for session 2635be7a-1607-4e1c-a973-7d0fbcc987a7, skipping enqueue (backend/api/features/chat/routes.py:1073) [TIMING] Session already completed, adding StreamFinish (backend/copilot/stream_registry.py:560).
4	Distinct user clicks, same session/text → fresh UUID per click	PASS	Click 1 sent ed2ca707-..., click 2 sent c0f70768-... (different UUIDs, both reflected in the persisted session message list). Two independent assistant turns ran, each with its own assistant message id (8995c86e... and faf7d821...).
5	Race-fix sanity: a normal send still enqueues + completes after the second is_turn_in_flight was removed	PASS	Sent "Race-fix sanity: please reply with the word OK only." with fresh message_id=6c53ab5f-705b-4d8c-9d09-e2f2d509ccfb, status 200, assistant streamed "OK".

Backend log excerpt for the PK-collision short-circuit

INFO  [STREAM] Duplicate message detected for session 2635be7a-1607-4e1c-a973-7d0fbcc987a7, skipping enqueue
  python_logger=backend.api.features.chat.routes
  source: /app/autogpt_platform/backend/backend/api/features/chat/routes.py:1073

INFO  [TIMING] Session already completed, adding StreamFinish
  python_logger=backend.copilot.stream_registry
  component=StreamRegistry session_status=completed
  source: /app/autogpt_platform/backend/backend/copilot/stream_registry.py:560

That is_duplicate_message branch is the one introduced in 93aa3dd25a (after append_and_save_message returns None on PK collision), and the second is_turn_in_flight check was removed in the same commit (matches the diff in routes.py:999-1037). No UniqueViolationError/P2002 lines surfaced in kubectl logs because the dedup is detected at the application layer (append_and_save_message returning None), not as a raised exception bubbling up.

Headline: PASS on all 5 scenarios.

[majdyz]

Chip-queue fix verified on dev preview (deploy 25179275328, HEAD 7204aef5d3)

Session: 802163ca-ab09-45af-be0e-29d41ac5c9df

Repro (jumbled scenario):

1. Initial: can you sleep for 2 seconds then 3 seconds
2. Chip 1 (during stream): oh do 5 secs sleep in between
3. Chip 2 (during stream): no actually 4
4. Chip 3 (delayed): okay nice

Results

Check	Result
User bubble for queued chip appears DURING streaming (not just after turn-end)	PASS
4 distinct user bubbles in chronological order after turn-end	PASS
No bubble shows joined \n\n content from another bubble	PASS
DB: 4 separate user rows with raw chip text, no envelopes on chip rows	PASS

Key signal: bubbles appeared DURING streaming as soon as backend drained each chip - confirmed via document.body.innerText snapshots taken ~1s after each chip submit. The previous broken behavior (bubbles only appearing post-hydration after turn-end) is gone.

DB verification (ChatMessage rows):

seq=  0 user       <memory_context>... (original send w/ envelope)
seq=  7 user       oh do 5 secs sleep in between
seq= 11 user       no actually 4
seq= 15 user       okay nice

Chip rows are raw text only - no envelopes, no \n\n-joined content. Each chip is interleaved with its own assistant + tool turn (seq 8-10, 12-14, 16-17).

Screenshots saved locally at test-results/PR-12948-chip-queue-final/{during-stream-1,during-stream-2,mid-stream-3,post-turn}.png.

LGTM - ready to merge.