[TMP] [TESTING] merge(preview): consolidated preview of 5 active PRs

[majdyz]

Summary

Temporary preview branch consolidating 5 active PRs for joint testing:

• fix(backend/copilot): robust context fallback — upload gate, gap-fill, token-budget compression #12782 fix(backend/copilot): robust context fallback — upload gate, gap-fill, token-budget compression
• feat(copilot): queue follow-up messages on busy sessions (UI + run_sub_session + AutoPilot block) #12737 feat(copilot): inject pending messages into running turns via Redis buffer
• fix(backend/executor): charge per LLM iteration and per tool call in OrchestratorBlock #12735 fix(backend/executor): charge per LLM iteration and per tool call in OrchestratorBlock
• feat(platform): subscription tier billing via Stripe Checkout #12727 feat(platform): subscription tier billing via Stripe Checkout
• feat(builder): AI chat panel for the flow builder #12699 feat(frontend/builder): add AI chat panel to flow builder

Do not merge — this branch is for preview/testing only. Each PR will be merged individually.

[github-actions]

Conflicts have been resolved! 🎉 A maintainer will review the pull request shortly.

[sentry]

Bug: The recursive call to stream_chat_completion_sdk in the auto-continue logic omits the model parameter, causing it to revert to the default model instead of the user's selection.
_{Severity: HIGH}

Suggested Fix

Forward the model parameter in the recursive call to stream_chat_completion_sdk within the auto-continue logic, similar to how the mode parameter is already being forwarded. This will ensure the user's selected model persists across auto-continued turns.

Prompt for AI Agent

Review the code at the location below. A potential bug has been identified by an AI
agent.
Verify if this is a real issue. If it is, propose a fix; if not, explain why it's not
valid.

Location: autogpt_platform/backend/backend/copilot/sdk/service.py#L3513-L3521

Potential issue: When a user selects a non-default model, such as "advanced" for Opus,
and queues multiple messages, the auto-continue feature recursively calls
`stream_chat_completion_sdk`. This recursive call omits the `model` parameter, causing
subsequent turns to fall back to the default model instead of respecting the user's
original choice. This leads to incorrect rate-limit quota charging, specifically
under-charging by approximately 80% for users who selected the premium model.

_{Did we get this right? 👍 / 👎 to inform future reviews.}

[coderabbitai]

Actionable comments posted: 16

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/components/ChatInput/__tests__/ChatInput.test.tsx (1)

22-35: ⚠️ Potential issue | 🔴 Critical

Fix the broken store mock before adding more assertions.

Line 25 still returns setCopilotMode: mockSetCopilotMode, but that symbol no longer exists after the rename to mockSetCopilotChatMode. This crashes render for the entire ChatInput suite, which matches the current CI failure.

Minimal fix

 vi.mock("@/app/(platform)/copilot/store", () => ({
   useCopilotUIStore: () => ({
     copilotMode: mockCopilotMode,
-    setCopilotMode: mockSetCopilotMode,
+    setCopilotMode: mockSetCopilotChatMode,
     copilotChatMode: mockCopilotMode,
     setCopilotChatMode: mockSetCopilotChatMode,
     copilotLlmModel: mockCopilotLlmModel,
     setCopilotLlmModel: mockSetCopilotLlmModel,

🤖 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/components/ChatInput/__tests__/ChatInput.test.tsx
around lines 22 - 35, The store mock references a nonexistent symbol: update the
mocked useCopilotUIStore return to use the renamed setter (replace the outdated
setCopilotMode reference with mockSetCopilotChatMode) so the mock aligns with
the new variable names (mockSetCopilotChatMode, mockCopilotMode,
mockSetCopilotLlmModel, etc.); ensure the object keys match the actual hooks
used in ChatInput (e.g., setCopilotChatMode / setCopilotLlmModel) so the test
render no longer throws due to a missing mock function.

♻️ Duplicate comments (5)

autogpt_platform/backend/backend/api/features/v1.py (1)

849-876: ⚠️ Potential issue | 🔴 Critical

Still cancel Stripe on FREE downgrades even when payments are disabled.

The payment_enabled == False branch writes FREE directly to the DB and returns. If a user already has an active Stripe subscription and that flag is later turned off, they will keep getting billed in Stripe while the app shows them as downgraded. This path still needs to attempt cancel_stripe_subscription(user_id) before the DB write, regardless of the feature flag.

🤖 Prompt for AI Agents

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

In `@autogpt_platform/backend/backend/api/features/v1.py` around lines 849 - 876,
For downgrades to SubscriptionTier.FREE ensure you always attempt to cancel any
existing Stripe subscription by calling cancel_stripe_subscription(user_id)
before writing the FREE tier to the DB, even when payment_enabled is False; move
or duplicate the cancellation logic used in the payment_enabled branch
(including the try/except stripe.StripeError logging and HTTPException behavior)
to run unconditionally for the FREE branch, and keep the existing fallback that
if cancel_stripe_subscription returns False then call
set_subscription_tier(user_id, tier) immediately (since webhook won’t fire).

autogpt_platform/backend/backend/data/credit.py (2)

1779-1807: ⚠️ Potential issue | 🔴 Critical

Retrying invoice.payment_failed can dead-end on the unique constraint.

If _add_transaction(... transaction_key=invoice_id) succeeds and stripe.Invoice.pay(invoice_id) then fails, Stripe retries the webhook. The second delivery hits UniqueViolationError, which isn't caught here, so the handler exits before retrying Invoice.pay. That leaves the balance already deducted while Stripe keeps retrying the unpaid invoice. Treat duplicate transaction_keys as “already deducted” and continue to Invoice.pay.

🤖 Prompt for AI Agents

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

In `@autogpt_platform/backend/backend/data/credit.py` around lines 1779 - 1807,
The handler must treat a UniqueViolation on the idempotency key as "already
deducted" and continue to mark the Stripe invoice paid; wrap the call to
credit_model._add_transaction (the call that sets transaction_key=invoice_id) in
a try/except that catches the DB unique-key duplicate error (e.g.,
UniqueViolationError / IntegrityError with SQLSTATE 23505) and, on that specific
error, proceed as if the debit succeeded (i.e., continue to call
run_in_threadpool(stripe.Invoice.pay, invoice_id)); keep all other exceptions
propagated or logged as before. Ensure you reference
credit_model._add_transaction, transaction_key/invoice_id, and
run_in_threadpool(stripe.Invoice.pay, invoice_id) when locating the change.

---

1440-1444: ⚠️ Potential issue | 🟠 Major

Don't create a Stripe customer from the “modify existing subscription” path.

modify_stripe_subscription_for_tier() is supposed to detect and update an existing paid subscription, but get_stripe_customer_id() eagerly creates one when stripe_customer_id is missing. For admin-granted paid tiers, a request that later falls back to Checkout—or even fails validation upstream—now leaves behind an orphan Stripe customer. Read the stored stripe_customer_id and return False when it's absent.

🧩 Minimal fix

-    customer_id = await get_stripe_customer_id(user_id)
+    user = await get_user_by_id(user_id)
+    if not user.stripe_customer_id:
+        return False
+    customer_id = user.stripe_customer_id

🤖 Prompt for AI Agents

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

In `@autogpt_platform/backend/backend/data/credit.py` around lines 1440 - 1444,
modify_stripe_subscription_for_tier currently calls
get_stripe_customer_id(user_id) which eagerly creates a Stripe customer; change
the flow to read the stored stripe customer id without creating one and return
False if it's missing. Concretely, replace the call to get_stripe_customer_id in
modify_stripe_subscription_for_tier with a read-only lookup of the persisted
stripe_customer_id (or add a non-creating parameter to get_stripe_customer_id)
and if the value is falsy immediately return False before calling
stripe.Subscription.list or creating any customer.

autogpt_platform/backend/backend/copilot/sdk/service.py (2)

3504-3522: ⚠️ Potential issue | 🟠 Major

Keep auto-continue iterative instead of recursively re-entering the stream.

A burst of queued follow-ups still creates one nested async-generator frame per drained batch here. That keeps the old stack-growth risk alive and is the same problem previously called out on this path.

🤖 Prompt for AI Agents

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

In `@autogpt_platform/backend/backend/copilot/sdk/service.py` around lines 3504 -
3522, The current code re-enters the stream by calling async for event in
stream_chat_completion_sdk(...) inside the existing stream handler, causing
nested async-generator frames; replace this recursive pattern with an iterative
loop: after the initial check on ended_with_stream_error, repeatedly call
drain_pending_safe(session_id, log_prefix) and if it returns texts, build
auto_message and then iterate over the async generator returned by
stream_chat_completion_sdk(session_id=..., message=auto_message, ...) yielding
each event; repeat the drain->stream cycle until drain_pending_safe returns
empty or ended_with_stream_error becomes true, ensuring you reference
ended_with_stream_error, drain_pending_safe, stream_chat_completion_sdk,
_auto_pending_texts and auto_message and avoid re-invoking the outer stream
handler recursively.

---

3513-3521: ⚠️ Potential issue | 🟠 Major

Propagate the selected model tier into auto-continued turns.

The handoff drops model, so queued follow-ups from an advanced turn silently fall back to default model selection. That changes both behavior and billing/rate-limit accounting mid-conversation.

Suggested fix

             async for event in stream_chat_completion_sdk(
                 session_id=session_id,
                 message=auto_message,
                 is_user_message=True,
                 user_id=user_id,
                 file_ids=None,
                 permissions=permissions,
                 mode=mode,
+                model=model,
             ):
                 yield event

🤖 Prompt for AI Agents

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

In `@autogpt_platform/backend/backend/copilot/sdk/service.py` around lines 3513 -
3521, The auto-continued handoff call to stream_chat_completion_sdk is dropping
the selected model (causing advanced turns to fall back to defaults); update the
call site in service.py where stream_chat_completion_sdk(...) is invoked (the
async for loop that passes session_id, message=auto_message, is_user_message,
user_id, file_ids, permissions, mode) to also pass through the current model
variable (e.g., model=model) and ensure the stream_chat_completion_sdk signature
(and any intermediate callers) accept and forward a model argument so queued
follow-ups preserve the chosen model tier.

🧹 Nitpick comments (10)

autogpt_platform/frontend/src/app/(platform)/copilot/components/PulseChips/usePulseChips.ts (1)

6-21: The memo here isn’t buying anything.

This is just a cheap map over a tiny array, so the extra useMemo adds noise and bookkeeping for no real win.

♻️ Inline the mapping

-import { useMemo } from "react";
-
 export function usePulseChips(): PulseChipData[] {
   const { agents } = useLibraryAgents();

   const sitrepItems = useSitrepItems(agents, 5);

-  return useMemo(() => {
-    return sitrepItems.map((item) => ({
-      id: item.id,
-      agentID: item.agentID,
-      name: item.agentName,
-      status: item.status,
-      shortMessage: item.message,
-    }));
-  }, [sitrepItems]);
+  return sitrepItems.map((item) => ({
+    id: item.id,
+    agentID: item.agentID,
+    name: item.agentName,
+    status: item.status,
+    shortMessage: item.message,
+  }));
 }

As per coding guidelines, "Do not use useCallback or useMemo unless asked to optimise a given function".

🤖 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/components/PulseChips/usePulseChips.ts
around lines 6 - 21, The useMemo wrapper in usePulseChips is unnecessary; remove
useMemo and return the mapped array directly: call useLibraryAgents() and
useSitrepItems(agents, 5) as before, then immediately return
sitrepItems.map(item => ({ id: item.id, agentID: item.agentID, name:
item.agentName, status: item.status, shortMessage: item.message })); keep the
function signature usePulseChips(): PulseChipData[] and the same property names
to preserve types and behavior.

autogpt_platform/frontend/src/app/(platform)/library/components/SitrepItem/useSitrepItems.ts (1)

93-97: Drop the per-agent re-sort here.

groupByAgent() preserves the order from useGetV1ListAllExecutions, so this extra sort() adds work on every agent without changing which failure/completion .find() will pick.

♻️ Simplify the recent-executions path

-  const recent = executions
-    .filter((e) => endedAfter(e, cutoff))
-    .sort((a, b) => toEndTime(b) - toEndTime(a));
+  const recent = executions.filter((e) => endedAfter(e, cutoff));

Also remove toEndTime from the import list once this sort is gone.

Based on learnings, useGetV1ListAllExecutions already returns executions in reverse-chronological order (most recent first).

🤖 Prompt for AI Agents

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

In
`@autogpt_platform/frontend/src/app/`(platform)/library/components/SitrepItem/useSitrepItems.ts
around lines 93 - 97, The per-agent re-sort is redundant and costly: remove the
.sort((a, b) => toEndTime(b) - toEndTime(a)) from the recent variable (which
currently filters via endedAfter and assigns to recent) so recent preserves the
reverse-chronological order returned by useGetV1ListAllExecutions; also remove
toEndTime from the import list and ensure groupByAgent() continues to consume
recent as-is (keeping endedAfter and cutoff logic intact).

autogpt_platform/frontend/src/app/(platform)/library/components/AgentBriefingPanel/BriefingTabContent.tsx (1)

1-345: Consider splitting this file as it grows.

At ~345 lines, this file exceeds the ~200 line guideline. The sub-components (UsageSection, ExecutionListSection, AgentListSection, UsageFooter, UsageMeter, EmptyMessage) could be extracted into a local components/ folder if the file continues to grow.

As per coding guidelines: "Keep files under ~200 lines; extract sub-components or hooks into their own files when a file grows beyond this."

🤖 Prompt for AI Agents

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

In
`@autogpt_platform/frontend/src/app/`(platform)/library/components/AgentBriefingPanel/BriefingTabContent.tsx
around lines 1 - 345, The file is over the ~200-line guideline; extract the
large sub-components into separate files under a new local components/ folder
(e.g., UsageSection.tsx, ExecutionListSection.tsx, AgentListSection.tsx,
UsageFooter.tsx, UsageMeter.tsx, EmptyMessage.tsx) by moving the corresponding
functions (UsageSection, ExecutionListSection, AgentListSection, UsageFooter,
UsageMeter, EmptyMessage and any constants they need like MAX_VISIBLE and
TAB_STATUS_LABEL) into those modules, export them, and update BriefingTabContent
to import those components and any hooks/props (e.g., useGetV2GetCopilotUsage,
useSitrepItems, useResetRateLimit, useCredits) to preserve behavior; ensure all
types (CoPilotUsageStatus, LibraryAgent, AgentStatusFilter) are imported where
needed and keep default behavior and props unchanged.

autogpt_platform/frontend/src/app/(platform)/library/components/AgentBriefingPanel/StatsGrid.tsx (1)

14-65: Unused color property in TILES configuration.

Each tile in the TILES array defines a color property (e.g., "text-zinc-700", "text-blue-600"), but this property is never used in the render logic. Either apply the color to the value text or remove the unused property.

🧹 Option A: Apply the color to the value

-            <Text variant="h4">{value}</Text>
+            <Text variant="h4" className={tile.color}>{value}</Text>

🧹 Option B: Remove the unused property

 const TILES: {
   label: string;
   key: keyof FleetSummary;
   format?: (v: number) => string;
   filter: AgentStatusFilter;
   emoji: string;
-  color: string;
 }[] = [
   {
     label: "Spent this month",
     key: "monthlySpend",
     format: (v) => `$${v.toLocaleString()}`,
     filter: "all",
     emoji: "💵",
-    color: "text-zinc-700",
   },
   // ... remove color from all other tiles

🤖 Prompt for AI Agents

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

In
`@autogpt_platform/frontend/src/app/`(platform)/library/components/AgentBriefingPanel/StatsGrid.tsx
around lines 14 - 65, The TILES array defines a color property that isn't used;
update the StatsGrid component's render where each tile is mapped (referencing
the TILES constant) to apply tile.color as a CSS class on the value element (the
rendered numeric/string for each tile) so the color is used, ensuring you
preserve existing classNames and formatting (e.g., combine with any existing
classes via string interpolation or classnames utility); alternatively, if you
prefer not to style values, remove the color field from each object in TILES and
any related typings to eliminate the unused property.

autogpt_platform/frontend/src/app/(platform)/profile/(user)/credits/components/SubscriptionTierSection/SubscriptionTierSection.tsx (1)

80-83: Avoid adding more dark: overrides in this component.

These new styles hardcode dark-mode behavior in several places, but the frontend guidelines route theme handling through the design system instead of per-component dark:* classes. Please switch these to neutral/theme-token styles exposed by the shared components. As per coding guidelines, "No dark: Tailwind classes — the design system handles dark mode."

Also applies to: 98-103, 123-126, 215-216, 248-251

🤖 Prompt for AI Agents

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

In
`@autogpt_platform/frontend/src/app/`(platform)/profile/(user)/credits/components/SubscriptionTierSection/SubscriptionTierSection.tsx
around lines 80 - 83, The component SubscriptionTierSection adds several `dark:`
Tailwind overrides (e.g., the <p role="alert"> alert, plus other spans/divs
around lines you modified) which violates the design system rule; remove all
`dark:*` classes in this component and replace them with the shared
design-system theme tokens or shared components (e.g., use the shared
Alert/Badge/Token classes or the exported theme token class names instead of
`dark:bg-...`, `dark:text-...`, `dark:border-...`) so theme handling is
delegated to the design system; update every instance you changed (the alert
paragraph and the other className usages noted in the comment) to use the
appropriate shared component or token class names and props exposed by the
shared UI library rather than per-component `dark:` overrides.

autogpt_platform/backend/backend/api/features/library/db.py (1)

96-100: Consider using %s placeholders for debug log statements.

Per coding guidelines, debug log statements should use %s placeholders for deferred interpolation to avoid string formatting overhead when debug logging is disabled.

♻️ Suggested change

-    logger.debug(
-        f"Fetching library agents for user_id={user_id}, "
-        f"search_term={repr(search_term)}, "
-        f"sort_by={sort_by}, page={page}, page_size={page_size}"
-    )
+    logger.debug(
+        "Fetching library agents for user_id=%s, search_term=%r, sort_by=%s, page=%s, page_size=%s",
+        user_id,
+        search_term,
+        sort_by,
+        page,
+        page_size,
+    )

As per coding guidelines: "Use %s for deferred interpolation in debug log statements for efficiency".

🤖 Prompt for AI Agents

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

In `@autogpt_platform/backend/backend/api/features/library/db.py` around lines 96
- 100, Change the debug log to use deferred interpolation with %s placeholders
instead of f-strings: update the logger.debug call that currently formats
f"Fetching library agents..." to use a format string with %s placeholders and
pass user_id, repr(search_term), sort_by, page, page_size as separate arguments
to logger.debug (refer to the existing logger.debug invocation and the variables
user_id, search_term, sort_by, page, page_size).

autogpt_platform/backend/backend/copilot/token_tracking_test.py (1)

213-234: Add one non-default multiplier forwarding test for stronger coverage.

This test now validates the default path (1.0), but a single explicit non-default case would better protect the new multiplier plumbing end-to-end.

🧪 Suggested additive test

 class TestRateLimitRecording:
@@
     async def test_calls_record_token_usage_when_user_id_present(self):
@@
         mock_record.assert_awaited_once_with(
             user_id="user-abc",
             prompt_tokens=100,
             completion_tokens=50,
             cache_read_tokens=1000,
             cache_creation_tokens=200,
             model_cost_multiplier=1.0,
         )
+
+    `@pytest.mark.asyncio`
+    async def test_forwards_non_default_model_cost_multiplier(self):
+        mock_record = AsyncMock()
+        with patch(
+            "backend.copilot.token_tracking.record_token_usage",
+            new=mock_record,
+        ):
+            await persist_and_record_usage(
+                session=None,
+                user_id="user-abc",
+                prompt_tokens=100,
+                completion_tokens=50,
+                model_cost_multiplier=5.0,
+            )
+        mock_record.assert_awaited_once_with(
+            user_id="user-abc",
+            prompt_tokens=100,
+            completion_tokens=50,
+            cache_read_tokens=0,
+            cache_creation_tokens=0,
+            model_cost_multiplier=5.0,
+        )

🤖 Prompt for AI Agents

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

In `@autogpt_platform/backend/backend/copilot/token_tracking_test.py` around lines
213 - 234, Add a new async test in token_tracking_test.py that patches
backend.copilot.token_tracking.record_token_usage (like the existing test) but
calls persist_and_record_usage with a non-default model_cost_multiplier (e.g.,
1.5) and asserts record_token_usage was awaited once with the same user_id,
token counts, and model_cost_multiplier set to that non-default value; reference
the existing test_calls_record_token_usage_when_user_id_present and the
persist_and_record_usage/record_token_usage symbols to place and implement the
new test.

autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatInput/components/ModelToggleButton.tsx (1)

5-5: Consider sourcing CopilotLlmModel from generated OpenAPI types to avoid contract drift.

Using a store-local enum for a request contract type can diverge from backend schema over time; prefer a generated API type as the source of truth and re-export from the store if needed.

Based on learnings: “Prefer using generated OpenAPI types from '@/app/api/generated/' for payloads defined in openapi.json … avoid re-implementing types when a generated type is available.”

🤖 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/components/ChatInput/components/ModelToggleButton.tsx
at line 5, The file imports a local store type CopilotLlmModel which risks
drifting from the backend contract; replace that import with the generated
OpenAPI type (e.g., import the corresponding enum/type from
'@/app/api/__generated__' and use it in ModelToggleButton and any consumers), or
update the store to re-export the generated type so callers keep the same import
path; update the ModelToggleButton.tsx import to the generated type, adjust any
type names if needed, and run typechecks/build to ensure the new generated type
matches usage.

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

56-57: Rename these symbols to LLM, not Llm.

CopilotLlmModel / copilotLlmModel / setCopilotLlmModel break the frontend acronym-casing rule and will spread that inconsistency to every caller. Please switch this family to CopilotLLMModel / copilotLLMModel / setCopilotLLMModel before the new API settles.

As per coding guidelines, "Fully capitalize acronyms in symbols, e.g. graphID, useBackendAPI."

Also applies to: 140-145

🤖 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/store.ts around lines
56 - 57, Rename the Llm-acronym symbols to use uppercase LLM: change the type
CopilotLlmModel -> CopilotLLMModel, the state variable copilotLlmModel ->
copilotLLMModel, and the setter setCopilotLlmModel -> setCopilotLLMModel (and
any other occurrences from this family, including the ones referenced around the
140-145 region). Update all declarations, exports, imports, usages, and related
JSDoc/comments to the new names so callers and selectors compile cleanly and
maintain the frontend acronym-casing rule.

autogpt_platform/frontend/src/app/(platform)/build/components/BuilderChatPanel/useSessionManager.ts (1)

63-72: Please remove these eslint-disable escapes.

These effects are now core session-lifecycle code. Keeping react-hooks/exhaustive-deps suppressed here makes future changes much harder to reason about; it would be better to restructure the guards/refs so the real dependency set can be declared explicitly.

As per coding guidelines, "No linter suppressors (// @ts-ignore``, // eslint-disable) — fix the actual issue."

Also applies to: 123-130

🤖 Prompt for AI Agents

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

In
`@autogpt_platform/frontend/src/app/`(platform)/build/components/BuilderChatPanel/useSessionManager.ts
around lines 63 - 72, Remove the eslint-disable comment and make the effect
dependencies explicit in useSessionManager: update the useEffect that currently
references flowID, graphSessionCache, setSessionId, setSessionError,
isCreatingSessionRef, and hasSentSeedMessageRef so the dependency array includes
stable references (e.g., flowID and graphSessionCache) and ensure setters and
refs are stable (do not include refs in deps; useRef values need not be listed).
If any referenced functions/objects (like graphSessionCache or setters) are not
stable, memoize them (useMemo/useCallback) or derive their values inside the
effect to avoid stale closures. Apply the same change to the other effect at the
block around lines 123-130 so both effects declare precise dependencies instead
of disabling react-hooks/exhaustive-deps.

🤖 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/api/features/chat/routes_test.py`:
- Around line 1099-1101: The test function
test_stream_chat_accepts_exactly_max_length_message uses the wrong pytest-mock
type; replace the incorrect annotation pytest_mock.MockFixture with
pytest_mock.MockerFixture in the function signature so the test uses the correct
MockerFixture type (consistent with other tests) and avoids AttributeError at
collection time.
- Around line 202-205: The test is patching the wrong symbol: change the mock
target from backend.api.features.chat.routes.get_or_create_workspace to
backend.data.workspace.get_or_create_workspace so you mock where the function is
used via resolve_workspace_files; update the mocker.patch call that currently
references get_or_create_workspace in routes to instead patch
backend.data.workspace.get_or_create_workspace (matching the other workspace
mocks) so all workspace mocks are consistent.

In `@autogpt_platform/backend/backend/api/features/chat/routes.py`:
- Around line 909-914: The dedup lock acquired by acquire_dedup_lock(session_id,
original_message, sanitized_file_ids) can be leaked if any awaited setup step
throws before event_generator() is returned; ensure you release the dedup key
immediately on any setup failure by calling the existing dedup-release path (the
same cleanup used on normal finish/error) before re-raising or returning an
error response, i.e., wrap the setup sequence after acquire_dedup_lock in
try/except/finally (or explicitly release and then raise) so dedup_lock is
released on exceptions; apply the identical fix to the other code path that also
calls acquire_dedup_lock (the second acquire_dedup_lock usage) so both
acquisition sites free the lock on setup failure.
- Around line 906-914: The dedup lock key currently only uses original_message
and sanitized_file_ids, causing requests that differ by request.context,
request.mode, or request.model to be incorrectly deduplicated; update the calls
to acquire_dedup_lock (the invocation that sets dedup_lock and the similar call
at the later block) to include request.context, request.mode, and request.model
(or their normalized equivalents) alongside session_id, original_message, and
sanitized_file_ids so the fingerprint covers context, mode, and model; ensure
the acquire_dedup_lock implementation and any hashing/fingerprint helper consume
these new fields in the same order/format used by the two call sites.

In `@autogpt_platform/backend/backend/copilot/message_dedup_test.py`:
- Around line 33-44: Update the test_acquire_returns_lock_on_first_request to
assert that the Redis set call uses NX and TTL: when mocking Redis via
_patch_redis and calling acquire_dedup_lock("sess-1", "hello", None), verify
mock_redis.set was called with a key that starts with f"{_KEY_PREFIX}:sess-1:"
and that the call included nx=True and ex=30 (i.e., ensure the set invocation
from acquire_dedup_lock uses NX and a 30s TTL to enforce the dedup idempotency
contract).

In `@autogpt_platform/backend/backend/copilot/message_dedup.py`:
- Around line 25-39: The current _DedupLock implementation is unsafe across TTL
expiry because release() blindly deletes _key and can remove a newer lock;
modify lock creation to store a unique token (e.g., UUID) as the value at _key
and update the lock API (class _DedupLock) to keep that token, change release()
to perform a compare-and-delete (delete only if stored value equals the token)
using an atomic Redis script/command, and add a keepalive/extend method (or have
the lock refresh TTL periodically while held) that only updates the TTL for the
key if the token still matches; apply the same compare-and-delete token pattern
to the other lock usage referenced around lines 63-71.

In `@autogpt_platform/backend/backend/copilot/rate_limit.py`:
- Around line 341-343: The code computes total using a clamped multiplier but
logs the raw model_cost_multiplier; update the calculation to assign the clamped
value to a named variable (e.g., effective_multiplier = max(1.0,
model_cost_multiplier)), use effective_multiplier in the total calculation
((weighted_input + completion_tokens) * effective_multiplier) and change the log
statements to print effective_multiplier instead of model_cost_multiplier; apply
the same change to the other similar block referenced around lines 351-357 so
all telemetry shows the actual multiplier used.

In `@autogpt_platform/backend/backend/copilot/sdk/service.py`:
- Around line 718-725: The per-request override currently passes the hyphenated
identifier "anthropic/claude-opus-4-6" to _normalize_model_name, which prevents
OpenRouter-aware normalization; change the string passed in the model ==
"advanced" branch to the OpenRouter-standard dot form
"anthropic/claude-opus-4.6" so _normalize_model_name can correctly adapt routing
when config.openrouter_active is true (the rest of the block using session_id,
logger.info and returning sdk_model and _OPUS_COST_MULTIPLIER remains
unchanged).
- Around line 2575-2578: The code currently computes model_cost_multiplier via
_resolve_model_and_multiplier before the SDK may switch models via its fallback
logic; update the flow so the actual model used is captured when a fallback is
activated and used when persisting usage: when
fallback_model_activated_per_attempt is set, record the chosen fallback model
name into the request context (or a local variable) and recompute or look up the
model_cost_multiplier for that actual model immediately before calling
persist_and_record_usage; ensure persist_and_record_usage receives the actual
model name (not the pre-resolved one) and its corresponding multiplier so rate
limits and DB records reflect the model that actually ran.

In
`@autogpt_platform/frontend/src/app/`(platform)/build/components/BuilderChatPanel/__tests__/useBuilderChatPanel.test.ts:
- Around line 1271-1288: The test currently only checks that the seed prefix is
present in the first request but doesn't assert that the original user message
("hello") is preserved; update the first-request assertion for
ctorArg.prepareSendMessagesRequest({ messages }) (req) to also assert the
request body includes the original user text (e.g., expect(req).toMatchObject({
body: { is_user_message: true }, headers: { Authorization: "Bearer tok" } }) and
additionally assert the message string contains both the seed prefix and "hello"
or equals the expected concatenated string) so the test fails if "hello" is
dropped while keeping the existing second-request assertion for req2 unchanged.

In
`@autogpt_platform/frontend/src/app/`(platform)/build/components/BuilderChatPanel/useSessionManager.ts:
- Around line 148-168: After constructing and possibly mutating messageText (via
extractTextFromParts and buildSeedPrompt) ensure you clamp the final messageText
to StreamChatRequest.message's maxLength (64000) before returning the body;
specifically, in the block that sets body.message, replace the raw messageText
with a truncated version (e.g., messageText.slice(0, 64000)) so seeded first
turns produced by buildSeedPrompt cannot exceed the 64k limit and cause a 422.

In `@autogpt_platform/frontend/src/app/`(platform)/copilot/useCopilotPage.ts:
- Around line 268-295: When isInFlight is true we currently pass the full
trimmed text into postV2QueuePendingMessage which will 422 when message > 32000
chars; before calling postV2QueuePendingMessage (and before adding to
setQueuedMessages) clamp/validate trimmed to 32000 chars
(QueuePendingMessageRequest.message maxLength = 32000) — e.g., compute a
truncatedMessage = trimmed.slice(0, 32000) and use that for
postV2QueuePendingMessage(sessionId, { message: truncatedMessage }) and for
setQueuedMessages((prev) => [...prev, truncatedMessage]); optionally surface a
toast if truncation occurred.

In
`@autogpt_platform/frontend/src/app/`(platform)/library/components/LibraryAgentCard/LibraryAgentCard.tsx:
- Around line 158-165: The error tooltip is currently only hover-reachable
because TooltipTrigger wraps a plain div (card), so keyboard users can't access
statusInfo.lastError; update LibraryAgentCard to attach the TooltipTrigger to a
focusable control instead (e.g., wrap the StatusBadge component or replace the
div trigger with a button or anchor) or alternatively render
statusInfo.lastError inline when hasError is true; ensure TooltipTrigger,
StatusBadge and the hasError/statusInfo.lastError conditional are updated so the
trigger is keyboard-focusable and the tooltip content remains the same.

In
`@autogpt_platform/frontend/src/app/`(platform)/library/hooks/useAgentStatus.ts:
- Around line 96-123: The hook currently computes concrete statuses using
executions ?? [], which yields misleading idle states while the query is loading
or errored; update useAgentStatusMap to read isSuccess (from
useGetV1ListAllExecutions) and only run the memoized computation when isSuccess
is true, otherwise return a neutral result (e.g., an empty Map or a Map with a
loading/unknown status) instead of deriving statuses from missing data; ensure
you stop using executions ?? [] and only pass executions into
computeAgentStatus(agent, agentExecs) when isSuccess is true.
- Around line 144-208: The hook useFleetSummary currently builds and returns an
all-zero FleetSummary while the executions query is still loading; update it to
detect the query status from useGetV1ListAllExecutions (use its isSuccess flag)
and return a distinct loading state instead of the zeroed summary. Specifically,
inside useFleetSummary, read isSuccess from useGetV1ListAllExecutions and if
isSuccess is false return an explicit loading sentinel (e.g., null or a
FleetSummary with a loading boolean) so callers can differentiate "loading" from
a genuine empty fleet; only compute the counts inside the useMemo when isSuccess
is true and executions is available. Ensure references to useFleetSummary,
useGetV1ListAllExecutions, and FleetSummary are updated accordingly.

In `@autogpt_platform/frontend/src/app/`(platform)/library/page.tsx:
- Around line 24-27: The statusFilter dropdown is showing counts derived from
fleetSummary (created from useLibraryAgents()) while the UI list is scoped by
activeTab, causing mismatched counts on non-"all" tabs like Favorites; update
the logic so the summary is computed from the same dataset the active tab uses
(e.g., derive a tabScopedAgents list from useLibraryAgents() filtered by
activeTab and pass that into useLibraryFleetSummary) or, alternatively, clear or
hide statusFilter (via setStatusFilter) when activeTab !== "all" so the counts
and filter behavior stay consistent with the displayed agents.

---

Outside diff comments:
In
`@autogpt_platform/frontend/src/app/`(platform)/copilot/components/ChatInput/__tests__/ChatInput.test.tsx:
- Around line 22-35: The store mock references a nonexistent symbol: update the
mocked useCopilotUIStore return to use the renamed setter (replace the outdated
setCopilotMode reference with mockSetCopilotChatMode) so the mock aligns with
the new variable names (mockSetCopilotChatMode, mockCopilotMode,
mockSetCopilotLlmModel, etc.); ensure the object keys match the actual hooks
used in ChatInput (e.g., setCopilotChatMode / setCopilotLlmModel) so the test
render no longer throws due to a missing mock function.

---

Duplicate comments:
In `@autogpt_platform/backend/backend/api/features/v1.py`:
- Around line 849-876: For downgrades to SubscriptionTier.FREE ensure you always
attempt to cancel any existing Stripe subscription by calling
cancel_stripe_subscription(user_id) before writing the FREE tier to the DB, even
when payment_enabled is False; move or duplicate the cancellation logic used in
the payment_enabled branch (including the try/except stripe.StripeError logging
and HTTPException behavior) to run unconditionally for the FREE branch, and keep
the existing fallback that if cancel_stripe_subscription returns False then call
set_subscription_tier(user_id, tier) immediately (since webhook won’t fire).

In `@autogpt_platform/backend/backend/copilot/sdk/service.py`:
- Around line 3504-3522: The current code re-enters the stream by calling async
for event in stream_chat_completion_sdk(...) inside the existing stream handler,
causing nested async-generator frames; replace this recursive pattern with an
iterative loop: after the initial check on ended_with_stream_error, repeatedly
call drain_pending_safe(session_id, log_prefix) and if it returns texts, build
auto_message and then iterate over the async generator returned by
stream_chat_completion_sdk(session_id=..., message=auto_message, ...) yielding
each event; repeat the drain->stream cycle until drain_pending_safe returns
empty or ended_with_stream_error becomes true, ensuring you reference
ended_with_stream_error, drain_pending_safe, stream_chat_completion_sdk,
_auto_pending_texts and auto_message and avoid re-invoking the outer stream
handler recursively.
- Around line 3513-3521: The auto-continued handoff call to
stream_chat_completion_sdk is dropping the selected model (causing advanced
turns to fall back to defaults); update the call site in service.py where
stream_chat_completion_sdk(...) is invoked (the async for loop that passes
session_id, message=auto_message, is_user_message, user_id, file_ids,
permissions, mode) to also pass through the current model variable (e.g.,
model=model) and ensure the stream_chat_completion_sdk signature (and any
intermediate callers) accept and forward a model argument so queued follow-ups
preserve the chosen model tier.

In `@autogpt_platform/backend/backend/data/credit.py`:
- Around line 1779-1807: The handler must treat a UniqueViolation on the
idempotency key as "already deducted" and continue to mark the Stripe invoice
paid; wrap the call to credit_model._add_transaction (the call that sets
transaction_key=invoice_id) in a try/except that catches the DB unique-key
duplicate error (e.g., UniqueViolationError / IntegrityError with SQLSTATE
23505) and, on that specific error, proceed as if the debit succeeded (i.e.,
continue to call run_in_threadpool(stripe.Invoice.pay, invoice_id)); keep all
other exceptions propagated or logged as before. Ensure you reference
credit_model._add_transaction, transaction_key/invoice_id, and
run_in_threadpool(stripe.Invoice.pay, invoice_id) when locating the change.
- Around line 1440-1444: modify_stripe_subscription_for_tier currently calls
get_stripe_customer_id(user_id) which eagerly creates a Stripe customer; change
the flow to read the stored stripe customer id without creating one and return
False if it's missing. Concretely, replace the call to get_stripe_customer_id in
modify_stripe_subscription_for_tier with a read-only lookup of the persisted
stripe_customer_id (or add a non-creating parameter to get_stripe_customer_id)
and if the value is falsy immediately return False before calling
stripe.Subscription.list or creating any customer.

---

Nitpick comments:
In `@autogpt_platform/backend/backend/api/features/library/db.py`:
- Around line 96-100: Change the debug log to use deferred interpolation with %s
placeholders instead of f-strings: update the logger.debug call that currently
formats f"Fetching library agents..." to use a format string with %s
placeholders and pass user_id, repr(search_term), sort_by, page, page_size as
separate arguments to logger.debug (refer to the existing logger.debug
invocation and the variables user_id, search_term, sort_by, page, page_size).

In `@autogpt_platform/backend/backend/copilot/token_tracking_test.py`:
- Around line 213-234: Add a new async test in token_tracking_test.py that
patches backend.copilot.token_tracking.record_token_usage (like the existing
test) but calls persist_and_record_usage with a non-default
model_cost_multiplier (e.g., 1.5) and asserts record_token_usage was awaited
once with the same user_id, token counts, and model_cost_multiplier set to that
non-default value; reference the existing
test_calls_record_token_usage_when_user_id_present and the
persist_and_record_usage/record_token_usage symbols to place and implement the
new test.

In
`@autogpt_platform/frontend/src/app/`(platform)/build/components/BuilderChatPanel/useSessionManager.ts:
- Around line 63-72: Remove the eslint-disable comment and make the effect
dependencies explicit in useSessionManager: update the useEffect that currently
references flowID, graphSessionCache, setSessionId, setSessionError,
isCreatingSessionRef, and hasSentSeedMessageRef so the dependency array includes
stable references (e.g., flowID and graphSessionCache) and ensure setters and
refs are stable (do not include refs in deps; useRef values need not be listed).
If any referenced functions/objects (like graphSessionCache or setters) are not
stable, memoize them (useMemo/useCallback) or derive their values inside the
effect to avoid stale closures. Apply the same change to the other effect at the
block around lines 123-130 so both effects declare precise dependencies instead
of disabling react-hooks/exhaustive-deps.

In
`@autogpt_platform/frontend/src/app/`(platform)/copilot/components/ChatInput/components/ModelToggleButton.tsx:
- Line 5: The file imports a local store type CopilotLlmModel which risks
drifting from the backend contract; replace that import with the generated
OpenAPI type (e.g., import the corresponding enum/type from
'@/app/api/__generated__' and use it in ModelToggleButton and any consumers), or
update the store to re-export the generated type so callers keep the same import
path; update the ModelToggleButton.tsx import to the generated type, adjust any
type names if needed, and run typechecks/build to ensure the new generated type
matches usage.

In
`@autogpt_platform/frontend/src/app/`(platform)/copilot/components/PulseChips/usePulseChips.ts:
- Around line 6-21: The useMemo wrapper in usePulseChips is unnecessary; remove
useMemo and return the mapped array directly: call useLibraryAgents() and
useSitrepItems(agents, 5) as before, then immediately return
sitrepItems.map(item => ({ id: item.id, agentID: item.agentID, name:
item.agentName, status: item.status, shortMessage: item.message })); keep the
function signature usePulseChips(): PulseChipData[] and the same property names
to preserve types and behavior.

In `@autogpt_platform/frontend/src/app/`(platform)/copilot/store.ts:
- Around line 56-57: Rename the Llm-acronym symbols to use uppercase LLM: change
the type CopilotLlmModel -> CopilotLLMModel, the state variable copilotLlmModel
-> copilotLLMModel, and the setter setCopilotLlmModel -> setCopilotLLMModel (and
any other occurrences from this family, including the ones referenced around the
140-145 region). Update all declarations, exports, imports, usages, and related
JSDoc/comments to the new names so callers and selectors compile cleanly and
maintain the frontend acronym-casing rule.

In
`@autogpt_platform/frontend/src/app/`(platform)/library/components/AgentBriefingPanel/BriefingTabContent.tsx:
- Around line 1-345: The file is over the ~200-line guideline; extract the large
sub-components into separate files under a new local components/ folder (e.g.,
UsageSection.tsx, ExecutionListSection.tsx, AgentListSection.tsx,
UsageFooter.tsx, UsageMeter.tsx, EmptyMessage.tsx) by moving the corresponding
functions (UsageSection, ExecutionListSection, AgentListSection, UsageFooter,
UsageMeter, EmptyMessage and any constants they need like MAX_VISIBLE and
TAB_STATUS_LABEL) into those modules, export them, and update BriefingTabContent
to import those components and any hooks/props (e.g., useGetV2GetCopilotUsage,
useSitrepItems, useResetRateLimit, useCredits) to preserve behavior; ensure all
types (CoPilotUsageStatus, LibraryAgent, AgentStatusFilter) are imported where
needed and keep default behavior and props unchanged.

In
`@autogpt_platform/frontend/src/app/`(platform)/library/components/AgentBriefingPanel/StatsGrid.tsx:
- Around line 14-65: The TILES array defines a color property that isn't used;
update the StatsGrid component's render where each tile is mapped (referencing
the TILES constant) to apply tile.color as a CSS class on the value element (the
rendered numeric/string for each tile) so the color is used, ensuring you
preserve existing classNames and formatting (e.g., combine with any existing
classes via string interpolation or classnames utility); alternatively, if you
prefer not to style values, remove the color field from each object in TILES and
any related typings to eliminate the unused property.

In
`@autogpt_platform/frontend/src/app/`(platform)/library/components/SitrepItem/useSitrepItems.ts:
- Around line 93-97: The per-agent re-sort is redundant and costly: remove the
.sort((a, b) => toEndTime(b) - toEndTime(a)) from the recent variable (which
currently filters via endedAfter and assigns to recent) so recent preserves the
reverse-chronological order returned by useGetV1ListAllExecutions; also remove
toEndTime from the import list and ensure groupByAgent() continues to consume
recent as-is (keeping endedAfter and cutoff logic intact).

In
`@autogpt_platform/frontend/src/app/`(platform)/profile/(user)/credits/components/SubscriptionTierSection/SubscriptionTierSection.tsx:
- Around line 80-83: The component SubscriptionTierSection adds several `dark:`
Tailwind overrides (e.g., the <p role="alert"> alert, plus other spans/divs
around lines you modified) which violates the design system rule; remove all
`dark:*` classes in this component and replace them with the shared
design-system theme tokens or shared components (e.g., use the shared
Alert/Badge/Token classes or the exported theme token class names instead of
`dark:bg-...`, `dark:text-...`, `dark:border-...`) so theme handling is
delegated to the design system; update every instance you changed (the alert
paragraph and the other className usages noted in the comment) to use the
appropriate shared component or token class names and props exposed by the
shared UI library rather than per-component `dark:` overrides.

🪄 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: 5091c22c-0748-4ec8-afab-9e229fa24b76

📥 Commits

Reviewing files that changed from the base of the PR and between 54bf060 and 6e56c29.

⛔ Files ignored due to path filters (1)

• autogpt_platform/frontend/pnpm-lock.yaml is excluded by !**/pnpm-lock.yaml

📒 Files selected for processing (114)

• autogpt_platform/backend/backend/api/features/chat/routes.py
• autogpt_platform/backend/backend/api/features/chat/routes_test.py
• autogpt_platform/backend/backend/api/features/library/db.py
• autogpt_platform/backend/backend/api/features/library/db_test.py
• autogpt_platform/backend/backend/api/features/library/model.py
• autogpt_platform/backend/backend/api/features/library/model_test.py
• autogpt_platform/backend/backend/api/features/subscription_routes_test.py
• autogpt_platform/backend/backend/api/features/v1.py
• autogpt_platform/backend/backend/copilot/baseline/service.py
• autogpt_platform/backend/backend/copilot/baseline/service_unit_test.py
• autogpt_platform/backend/backend/copilot/config.py
• autogpt_platform/backend/backend/copilot/executor/processor.py
• autogpt_platform/backend/backend/copilot/executor/utils.py
• autogpt_platform/backend/backend/copilot/message_dedup.py
• autogpt_platform/backend/backend/copilot/message_dedup_test.py
• autogpt_platform/backend/backend/copilot/pending_message_helpers.py
• autogpt_platform/backend/backend/copilot/pending_message_helpers_test.py
• autogpt_platform/backend/backend/copilot/pending_messages.py
• autogpt_platform/backend/backend/copilot/pending_messages_test.py
• autogpt_platform/backend/backend/copilot/prompt_cache_test.py
• autogpt_platform/backend/backend/copilot/prompting.py
• autogpt_platform/backend/backend/copilot/prompting_test.py
• autogpt_platform/backend/backend/copilot/rate_limit.py
• autogpt_platform/backend/backend/copilot/sdk/p0_guardrails_test.py
• autogpt_platform/backend/backend/copilot/sdk/query_builder_test.py
• autogpt_platform/backend/backend/copilot/sdk/retry_scenarios_test.py
• autogpt_platform/backend/backend/copilot/sdk/service.py
• autogpt_platform/backend/backend/copilot/sdk/service_helpers_test.py
• autogpt_platform/backend/backend/copilot/sdk/service_test.py
• autogpt_platform/backend/backend/copilot/service.py
• autogpt_platform/backend/backend/copilot/stream_registry.py
• autogpt_platform/backend/backend/copilot/stream_registry_test.py
• autogpt_platform/backend/backend/copilot/token_tracking.py
• autogpt_platform/backend/backend/copilot/token_tracking_test.py
• autogpt_platform/backend/backend/data/credit.py
• autogpt_platform/backend/backend/data/credit_subscription_test.py
• autogpt_platform/backend/backend/data/platform_cost_test.py
• autogpt_platform/backend/backend/data/workspace.py
• autogpt_platform/backend/backend/util/cache.py
• autogpt_platform/backend/backend/util/cache_test.py
• autogpt_platform/backend/backend/util/feature_flag.py
• autogpt_platform/backend/test/copilot/dry_run_loop_test.py
• autogpt_platform/frontend/package.json
• autogpt_platform/frontend/src/app/(platform)/build/components/BuilderChatPanel/BuilderChatPanel.tsx
• autogpt_platform/frontend/src/app/(platform)/build/components/BuilderChatPanel/__tests__/BuilderChatPanel.test.tsx
• autogpt_platform/frontend/src/app/(platform)/build/components/BuilderChatPanel/__tests__/useBuilderChatPanel.test.ts
• autogpt_platform/frontend/src/app/(platform)/build/components/BuilderChatPanel/useActionParser.ts
• autogpt_platform/frontend/src/app/(platform)/build/components/BuilderChatPanel/useBuilderChatPanel.ts
• autogpt_platform/frontend/src/app/(platform)/build/components/BuilderChatPanel/useSessionManager.ts
• autogpt_platform/frontend/src/app/(platform)/build/components/BuilderChatPanel/useToolCallHandler.ts
• autogpt_platform/frontend/src/app/(platform)/build/components/FlowEditor/Flow/Flow.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/CopilotPage.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/__tests__/helpers.test.ts
• autogpt_platform/frontend/src/app/(platform)/copilot/__tests__/store.test.ts
• autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatContainer/ChatContainer.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatInput/ChatInput.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatInput/__tests__/ChatInput.test.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatInput/components/DryRunToggleButton.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatInput/components/ModeToggleButton.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatInput/components/ModelToggleButton.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatInput/components/__tests__/ModelToggleButton.test.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatInput/useVoiceRecording.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/components/EmptySession/EmptySession.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/components/EmptySession/components/EditNameDialog/EditNameDialog.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/components/PulseChips/PulseChips.module.css
• autogpt_platform/frontend/src/app/(platform)/copilot/components/PulseChips/PulseChips.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/components/PulseChips/types.ts
• autogpt_platform/frontend/src/app/(platform)/copilot/components/PulseChips/usePulseChips.ts
• autogpt_platform/frontend/src/app/(platform)/copilot/components/RateLimitResetDialog/RateLimitResetDialog.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/components/UsageLimits/UsagePanelContent.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/components/usageHelpers.ts
• autogpt_platform/frontend/src/app/(platform)/copilot/helpers.ts
• autogpt_platform/frontend/src/app/(platform)/copilot/store.ts
• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotPage.ts
• autogpt_platform/frontend/src/app/(platform)/copilot/useCopilotStream.ts
• autogpt_platform/frontend/src/app/(platform)/layout.tsx
• autogpt_platform/frontend/src/app/(platform)/library/__tests__/main.test.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/AgentBriefingPanel/AgentBriefingPanel.module.css
• autogpt_platform/frontend/src/app/(platform)/library/components/AgentBriefingPanel/AgentBriefingPanel.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/AgentBriefingPanel/BriefingTabContent.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/AgentBriefingPanel/StatsGrid.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/AgentFilterMenu/AgentFilterMenu.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/ContextualActionButton/ContextualActionButton.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/JumpBackIn/JumpBackIn.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/JumpBackIn/useJumpBackIn.ts
• autogpt_platform/frontend/src/app/(platform)/library/components/LibraryActionHeader/LibraryActionHeader.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/LibraryAgentCard/LibraryAgentCard.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/LibraryAgentList/LibraryAgentList.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/LibraryAgentList/useLibraryAgentList.ts
• autogpt_platform/frontend/src/app/(platform)/library/components/LibraryFolder/LibraryFolder.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/LibrarySubSection/LibrarySubSection.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/SitrepItem/SitrepItem.module.css
• autogpt_platform/frontend/src/app/(platform)/library/components/SitrepItem/SitrepItem.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/SitrepItem/SitrepList.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/SitrepItem/useSitrepItems.ts
• autogpt_platform/frontend/src/app/(platform)/library/components/StatusBadge/StatusBadge.tsx
• autogpt_platform/frontend/src/app/(platform)/library/hooks/executionHelpers.ts
• autogpt_platform/frontend/src/app/(platform)/library/hooks/useAgentStatus.ts
• autogpt_platform/frontend/src/app/(platform)/library/hooks/useLibraryFleetSummary.ts
• autogpt_platform/frontend/src/app/(platform)/library/page.tsx
• autogpt_platform/frontend/src/app/(platform)/library/types.ts
• autogpt_platform/frontend/src/app/(platform)/profile/(user)/credits/components/SubscriptionTierSection/SubscriptionTierSection.tsx
• autogpt_platform/frontend/src/app/(platform)/profile/(user)/credits/components/SubscriptionTierSection/__tests__/SubscriptionTierSection.test.tsx
• autogpt_platform/frontend/src/app/(platform)/profile/(user)/credits/components/SubscriptionTierSection/useSubscriptionTierSection.ts
• autogpt_platform/frontend/src/app/api/openapi.json
• autogpt_platform/frontend/src/app/layout.tsx
• autogpt_platform/frontend/src/components/AgentationDevtool.tsx
• autogpt_platform/frontend/src/contexts/AutoPilotBridgeContext.tsx
• autogpt_platform/frontend/src/lib/autogpt-server-api/client.ts
• autogpt_platform/frontend/src/services/feature-flags/use-get-flag.ts
• autogpt_platform/frontend/src/services/storage/local-storage.ts
• codecov.yml

💤 Files with no reviewable changes (3)

• autogpt_platform/frontend/src/app/(platform)/library/components/JumpBackIn/JumpBackIn.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/JumpBackIn/useJumpBackIn.ts
• autogpt_platform/frontend/src/lib/autogpt-server-api/client.ts

✅ Files skipped from review due to trivial changes (17)

• autogpt_platform/backend/test/copilot/dry_run_loop_test.py
• autogpt_platform/frontend/src/services/storage/local-storage.ts
• autogpt_platform/frontend/src/app/(platform)/library/components/LibraryActionHeader/LibraryActionHeader.tsx
• autogpt_platform/backend/backend/data/platform_cost_test.py
• codecov.yml
• autogpt_platform/frontend/package.json
• autogpt_platform/frontend/src/services/feature-flags/use-get-flag.ts
• autogpt_platform/frontend/src/app/(platform)/copilot/components/PulseChips/types.ts
• autogpt_platform/frontend/src/components/AgentationDevtool.tsx
• autogpt_platform/frontend/src/app/(platform)/library/components/AgentBriefingPanel/AgentBriefingPanel.module.css
• autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatMessagesContainer/tests/ChatMessagesContainer.test.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/components/usageHelpers.ts
• autogpt_platform/frontend/src/app/(platform)/copilot/components/PulseChips/PulseChips.module.css
• autogpt_platform/backend/backend/copilot/pending_messages_test.py
• autogpt_platform/frontend/src/app/(platform)/library/components/SitrepItem/SitrepItem.module.css
• autogpt_platform/backend/backend/copilot/pending_message_helpers_test.py
• autogpt_platform/frontend/src/app/(platform)/library/types.ts

🚧 Files skipped from review as they are similar to previous changes (18)

• autogpt_platform/backend/backend/util/feature_flag.py
• autogpt_platform/backend/backend/copilot/sdk/retry_scenarios_test.py
• autogpt_platform/backend/backend/copilot/service.py
• autogpt_platform/frontend/src/app/(platform)/build/components/FlowEditor/Flow/Flow.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/CopilotPage.tsx
• autogpt_platform/frontend/src/app/(platform)/build/components/BuilderChatPanel/tests/BuilderChatPanel.test.tsx
• autogpt_platform/backend/backend/util/cache_test.py
• autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatMessagesContainer/ChatMessagesContainer.tsx
• autogpt_platform/backend/backend/util/cache.py
• autogpt_platform/backend/backend/copilot/sdk/query_builder_test.py
• autogpt_platform/frontend/src/app/(platform)/profile/(user)/credits/components/SubscriptionTierSection/tests/SubscriptionTierSection.test.tsx
• autogpt_platform/frontend/src/app/(platform)/build/components/BuilderChatPanel/BuilderChatPanel.tsx
• autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatInput/ChatInput.tsx
• autogpt_platform/backend/backend/copilot/pending_message_helpers.py
• autogpt_platform/frontend/src/app/(platform)/copilot/components/ChatContainer/ChatContainer.tsx
• autogpt_platform/frontend/src/app/(platform)/profile/(user)/credits/components/SubscriptionTierSection/useSubscriptionTierSection.ts
• autogpt_platform/backend/backend/copilot/pending_messages.py
• autogpt_platform/frontend/src/app/api/openapi.json

[coderabbitai]

⚠️ Potential issue | 🟠 Major

🧩 Analysis chain

🏁 Script executed:

#!/bin/bash
# Verify how get_or_create_workspace is imported and used in routes.py
rg -n "get_or_create_workspace" autogpt_platform/backend/backend/api/features/chat/routes.py

Repository: Significant-Gravitas/AutoGPT

Length of output: 54

---

🏁 Script executed:

#!/bin/bash
# Check if get_or_create_workspace is lazily imported or used anywhere in routes.py
# Search more broadly including imports
head -50 autogpt_platform/backend/backend/api/features/chat/routes.py | cat -n

# Also search for any conditional/lazy imports containing this name
rg "get_or_create_workspace" autogpt_platform/backend/backend/api/features/chat/routes.py -A 3 -B 3 2>/dev/null || echo "Not found in routes.py"

# Count lines in routes.py to see if line 1897 is even in this file
wc -l autogpt_platform/backend/backend/api/features/chat/routes.py

# Search in test file for line 1897 context
rg -n "get_or_create_workspace" autogpt_platform/backend/backend/api/features/chat/routes_test.py | head -20

Repository: Significant-Gravitas/AutoGPT

Length of output: 2401

---

🏁 Script executed:

#!/bin/bash
# Search for "workspace" in routes.py to find how it's used (module reference or lazy import)
rg "workspace" autogpt_platform/backend/backend/api/features/chat/routes.py -n

# Look at the test file around line 1897 to see what test case uses that old patch target
sed -n '1890,1910p' autogpt_platform/backend/backend/api/features/chat/routes_test.py | cat -n

# Search for lazy imports of get_or_create_workspace
rg "from backend.data.workspace import" autogpt_platform/backend/backend/api/features/chat/routes.py

Repository: Significant-Gravitas/AutoGPT

Length of output: 1348

---

🏁 Script executed:

#!/bin/bash
# Search more comprehensively for get_or_create_workspace usage in routes.py
rg "get_or_create_workspace" autogpt_platform/backend/backend/api/features/chat/routes.py -C 5

# If not found, try searching across all backend files to see where it's actually used
if ! rg "get_or_create_workspace" autogpt_platform/backend/backend/api/features/chat/routes.py &>/dev/null; then
  echo "=== get_or_create_workspace NOT found in routes.py ==="
  echo ""
  echo "=== Checking where get_or_create_workspace IS defined and used ==="
  rg "def get_or_create_workspace" autogpt_platform/backend/ --type py
fi

# Get full context around line 1897 in test file
echo ""
echo "=== Context around line 1897 in routes_test.py ==="
sed -n '1880,1920p' autogpt_platform/backend/backend/api/features/chat/routes_test.py | cat -n

Repository: Significant-Gravitas/AutoGPT

Length of output: 2019

---

🏁 Script executed:

#!/bin/bash
# Search for any conditional/lazy imports of get_or_create_workspace in routes.py
# including inside functions or if blocks
rg "if.*workspace|import.*workspace" autogpt_platform/backend/backend/api/features/chat/routes.py -A 3 -B 1

# Check if line 1897 test is in the same test function or a different one
echo "=== Test function containing line 1897 ==="
sed -n '1870,1900p' autogpt_platform/backend/backend/api/features/chat/routes_test.py | cat -n

# Search entire routes_test.py to see all test functions that use get_or_create_workspace patch
echo ""
echo "=== All patch targets for get_or_create_workspace in routes_test.py ==="
rg "\.patch\(" autogpt_platform/backend/backend/api/features/chat/routes_test.py | grep -i "get_or_create_workspace" | cat

Repository: Significant-Gravitas/AutoGPT

Length of output: 1897

---

🏁 Script executed:

#!/bin/bash
# Search for lazy imports - look for import statements inside functions or conditional blocks
rg "import.*get_or_create_workspace|from.*import.*get_or_create_workspace" autogpt_platform/backend/backend/api/features/chat/routes.py -B 5 -A 5

# Also check if workspace module is imported and used as workspace.get_or_create_workspace
rg "workspace\.get_or_create_workspace" autogpt_platform/backend/backend/api/features/chat/routes.py

# Look more broadly - check if the function name appears in any string or dynamic call
rg "get_or_create_workspace" autogpt_platform/backend/backend/api/features/chat/routes.py -i

# Let's check the exact test again - line 1897 and see the whole test function
echo "=== Full test function at line 1897 ==="
sed -n '1883,1950p' autogpt_platform/backend/backend/api/features/chat/routes_test.py | cat -n

Repository: Significant-Gravitas/AutoGPT

Length of output: 3071

---

🏁 Script executed:

#!/bin/bash
# Since get_or_create_workspace is not found in routes.py at all,
# let's check if the test is actually exercising code that needs this mock
# by looking at what the test does

# Check if the test route being called (/sessions/sess-hash/stream) actually uses get_or_create_workspace
echo "=== Searching for the stream endpoint handler in routes.py ==="
rg "def.*stream|@router.*stream" autogpt_platform/backend/backend/api/features/chat/routes.py -A 2 | head -30

# Let's search for file_ids handling since the test sends file_ids
rg "file_ids|resolve_workspace_files" autogpt_platform/backend/backend/api/features/chat/routes.py -B 3 -A 3 | head -40

# Check if maybe get_or_create_workspace was recently removed
git log --oneline --all -S "get_or_create_workspace" -- autogpt_platform/backend/backend/api/features/chat/routes.py 2>/dev/null | head -10 || echo "Git log search not available or no history"

Repository: Significant-Gravitas/AutoGPT

Length of output: 2304

---

🏁 Script executed:

#!/bin/bash
# Check if resolve_workspace_files or build_files_block call get_or_create_workspace
echo "=== Checking resolve_workspace_files and build_files_block implementations ==="
rg "def resolve_workspace_files|def build_files_block" autogpt_platform/backend/backend/data/workspace.py -A 20 | head -50

# Also check if get_or_create_workspace is called by those functions
rg "get_or_create_workspace" autogpt_platform/backend/backend/data/workspace.py -B 5 -A 5

Repository: Significant-Gravitas/AutoGPT

Length of output: 2376

---

Update line 1897 patch target to match other workspace mocks.

Line 1897 patches backend.api.features.chat.routes.get_or_create_workspace, but get_or_create_workspace is not imported in routes.py. It is only called indirectly via resolve_workspace_files() (line 85 import), which internally invokes it. Per the coding guideline to "mock where the symbol is used, not where it's defined", the patch target should be backend.data.workspace.get_or_create_workspace to match lines 203, 232, 270, and 899.

🤖 Prompt for AI Agents

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

In `@autogpt_platform/backend/backend/api/features/chat/routes_test.py` around
lines 202 - 205, The test is patching the wrong symbol: change the mock target
from backend.api.features.chat.routes.get_or_create_workspace to
backend.data.workspace.get_or_create_workspace so you mock where the function is
used via resolve_workspace_files; update the mocker.patch call that currently
references get_or_create_workspace in routes to instead patch
backend.data.workspace.get_or_create_workspace (matching the other workspace
mocks) so all workspace mocks are consistent.

[coderabbitai]

⚠️ Potential issue | 🔴 Critical

Typo: MockFixture should be MockerFixture.

pytest_mock.MockFixture does not exist. This will raise AttributeError at test collection time, preventing this test from running. The correct class is MockerFixture, consistent with all other tests in this file.

🐛 Proposed fix

 def test_stream_chat_accepts_exactly_max_length_message(
-    mocker: pytest_mock.MockFixture,
+    mocker: pytest_mock.MockerFixture,
 ):

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	def test_stream_chat_accepts_exactly_max_length_message(
	mocker: pytest_mock.MockFixture,
	):
	def test_stream_chat_accepts_exactly_max_length_message(
	mocker: pytest_mock.MockerFixture,
	):

🤖 Prompt for AI Agents

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

In `@autogpt_platform/backend/backend/api/features/chat/routes_test.py` around
lines 1099 - 1101, The test function
test_stream_chat_accepts_exactly_max_length_message uses the wrong pytest-mock
type; replace the incorrect annotation pytest_mock.MockFixture with
pytest_mock.MockerFixture in the function signature so the test uses the correct
MockerFixture type (consistent with other tests) and avoids AttributeError at
collection time.

[coderabbitai]

⚠️ Potential issue | 🟠 Major

Include context, mode, and model in the dedup fingerprint.

The lock key only hashes message text plus file IDs, but the queued executor payload now also depends on request.context, request.mode, and request.model. Two POSTs with the same text/files but different page context or model tier will be treated as duplicates even though they should execute differently.

Also applies to: 975-985

🤖 Prompt for AI Agents

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

In `@autogpt_platform/backend/backend/api/features/chat/routes.py` around lines
906 - 914, The dedup lock key currently only uses original_message and
sanitized_file_ids, causing requests that differ by request.context,
request.mode, or request.model to be incorrectly deduplicated; update the calls
to acquire_dedup_lock (the invocation that sets dedup_lock and the similar call
at the later block) to include request.context, request.mode, and request.model
(or their normalized equivalents) alongside session_id, original_message, and
sanitized_file_ids so the fingerprint covers context, mode, and model; ensure
the acquire_dedup_lock implementation and any hashing/fingerprint helper consume
these new fields in the same order/format used by the two call sites.

[coderabbitai]

⚠️ Potential issue | 🟠 Major

Release the dedup key if setup fails before the SSE generator starts.

From acquire_dedup_lock(...) to return StreamingResponse(...) there are multiple awaited setup steps. If one of them raises, event_generator().finally never runs, so retries for the same message are suppressed until the 30s TTL expires.

Based on learnings, the dedup key is intentionally retained on GeneratorExit but released on normal finish/error; setup failures before event_generator() currently hit neither path.

Also applies to: 1097-1100

🤖 Prompt for AI Agents

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

In `@autogpt_platform/backend/backend/api/features/chat/routes.py` around lines
909 - 914, The dedup lock acquired by acquire_dedup_lock(session_id,
original_message, sanitized_file_ids) can be leaked if any awaited setup step
throws before event_generator() is returned; ensure you release the dedup key
immediately on any setup failure by calling the existing dedup-release path (the
same cleanup used on normal finish/error) before re-raising or returning an
error response, i.e., wrap the setup sequence after acquire_dedup_lock in
try/except/finally (or explicitly release and then raise) so dedup_lock is
released on exceptions; apply the identical fix to the other code path that also
calls acquire_dedup_lock (the second acquire_dedup_lock usage) so both
acquisition sites free the lock on setup failure.

[coderabbitai]

⚠️ Potential issue | 🟡 Minor

Assert Redis NX + TTL on lock acquisition to protect idempotency contract.

This test should also verify set(..., nx=True, ex=30) so regressions in dedup lock semantics are caught.

Suggested test assertion update

     mock_redis.set.assert_called_once()
     key_arg = mock_redis.set.call_args.args[0]
     assert key_arg.startswith(f"{_KEY_PREFIX}:sess-1:")
+    assert mock_redis.set.call_args.kwargs.get("nx") is True
+    assert mock_redis.set.call_args.kwargs.get("ex") == 30

Based on learnings: “Redis idempotency dedup key ... uses NX with 30s TTL ... 30s TTL is the fallback for unhandled paths.”

🤖 Prompt for AI Agents

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

In `@autogpt_platform/backend/backend/copilot/message_dedup_test.py` around lines
33 - 44, Update the test_acquire_returns_lock_on_first_request to assert that
the Redis set call uses NX and TTL: when mocking Redis via _patch_redis and
calling acquire_dedup_lock("sess-1", "hello", None), verify mock_redis.set was
called with a key that starts with f"{_KEY_PREFIX}:sess-1:" and that the call
included nx=True and ex=30 (i.e., ensure the set invocation from
acquire_dedup_lock uses NX and a 30s TTL to enforce the dedup idempotency
contract).

[coderabbitai]

⚠️ Potential issue | 🟡 Minor

Clamp queued messages to the /pending limit before enqueueing.

When the turn is already in flight, Lines 284-285 send the full trimmed text to postV2QueuePendingMessage. That endpoint caps message at 32,000 chars, so oversized follow-ups fail there even though the normal /stream path accepts up to 64,000. Clamp or validate before enqueueing so this new queue flow doesn't degrade into a 422 + generic error toast for long messages.

Possible fix

       if (isInFlight) {
+        const MAX_QUEUED_MESSAGE_LENGTH = 32_000;
+
         // File attachments cannot be included in a queued pending message —
         // the queue API does not support file_ids.  Inform the user and bail.
         if (files && files.length > 0) {
@@
         // Queue the message into the pending buffer so it is picked up between
         // tool-call rounds by the currently running executor turn.
         try {
-          await postV2QueuePendingMessage(sessionId, { message: trimmed });
-          setQueuedMessages((prev) => [...prev, trimmed]);
+          const queuedMessage = trimmed.slice(0, MAX_QUEUED_MESSAGE_LENGTH);
+          await postV2QueuePendingMessage(sessionId, { message: queuedMessage });
+          setQueuedMessages((prev) => [...prev, queuedMessage]);
         } catch (err) {

Based on learnings, `QueuePendingMessageRequest.message maxLength = 32000 (matches PendingMessage.content)`.

📝 Committable suggestion

‼️ IMPORTANT
Carefully review the code before committing. Ensure that it accurately replaces the highlighted code, contains no missing lines, and has no issues with indentation. Thoroughly test & benchmark the code to ensure it meets the requirements.

Suggested change

	if (isInFlight) {
	// File attachments cannot be included in a queued pending message —
	// the queue API does not support file_ids. Inform the user and bail.
	if (files && files.length > 0) {
	toast({
	title: "Please wait to attach files",
	description:
	"File attachments can't be queued until the current response finishes.",
	variant: "destructive",
	});
	return;
	}
	// Queue the message into the pending buffer so it is picked up between
	// tool-call rounds by the currently running executor turn.
	try {
	await postV2QueuePendingMessage(sessionId, { message: trimmed });
	setQueuedMessages((prev) => [...prev, trimmed]);
	} catch (err) {
	toast({
	title: "Could not queue message",
	description: "Please wait for the current response to finish.",
	variant: "destructive",
	});
	throw err;
	}
	return;
	}
	if (isInFlight) {
	const MAX_QUEUED_MESSAGE_LENGTH = 32_000;
	// File attachments cannot be included in a queued pending message —
	// the queue API does not support file_ids. Inform the user and bail.
	if (files && files.length > 0) {
	toast({
	title: "Please wait to attach files",
	description:
	"File attachments can't be queued until the current response finishes.",
	variant: "destructive",
	});
	return;
	}
	// Queue the message into the pending buffer so it is picked up between
	// tool-call rounds by the currently running executor turn.
	try {
	const queuedMessage = trimmed.slice(0, MAX_QUEUED_MESSAGE_LENGTH);
	await postV2QueuePendingMessage(sessionId, { message: queuedMessage });
	setQueuedMessages((prev) => [...prev, queuedMessage]);
	} catch (err) {
	toast({
	title: "Could not queue message",
	description: "Please wait for the current response to finish.",
	variant: "destructive",
	});
	throw err;
	}
	return;
	}

🤖 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/useCopilotPage.ts
around lines 268 - 295, When isInFlight is true we currently pass the full
trimmed text into postV2QueuePendingMessage which will 422 when message > 32000
chars; before calling postV2QueuePendingMessage (and before adding to
setQueuedMessages) clamp/validate trimmed to 32000 chars
(QueuePendingMessageRequest.message maxLength = 32000) — e.g., compute a
truncatedMessage = trimmed.slice(0, 32000) and use that for
postV2QueuePendingMessage(sessionId, { message: truncatedMessage }) and for
setQueuedMessages((prev) => [...prev, truncatedMessage]); optionally surface a
toast if truncation occurred.

[coderabbitai]

⚠️ Potential issue | 🟡 Minor

Make the error tooltip keyboard reachable.

TooltipTrigger is wrapping a plain div, so statusInfo.lastError is effectively hover-only here. Keyboard users can't focus the trigger and won't be able to access the error details. Please anchor the tooltip to a focusable control in the card (for example the error StatusBadge) or expose the message inline for the error state.

🤖 Prompt for AI Agents

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

In
`@autogpt_platform/frontend/src/app/`(platform)/library/components/LibraryAgentCard/LibraryAgentCard.tsx
around lines 158 - 165, The error tooltip is currently only hover-reachable
because TooltipTrigger wraps a plain div (card), so keyboard users can't access
statusInfo.lastError; update LibraryAgentCard to attach the TooltipTrigger to a
focusable control instead (e.g., wrap the StatusBadge component or replace the
div trigger with a button or anchor) or alternatively render
statusInfo.lastError inline when hasError is true; ensure TooltipTrigger,
StatusBadge and the hasError/statusInfo.lastError conditional are updated so the
trigger is keyboard-focusable and the tooltip content remains the same.

[coderabbitai]

⚠️ Potential issue | 🟠 Major

Don’t derive concrete agent statuses before the executions query succeeds.

With executions ?? [], a pending or failed query makes every agent look idle/listening/scheduled, so cards can briefly show the wrong state and real failures disappear on fetch errors. Please gate this map on isSuccess or surface an explicit loading state instead of computing statuses from missing data.

Based on learnings, prefer checking isSuccess when gating on React Query results because data may be undefined in error/idle states.

🤖 Prompt for AI Agents

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

In `@autogpt_platform/frontend/src/app/`(platform)/library/hooks/useAgentStatus.ts
around lines 96 - 123, The hook currently computes concrete statuses using
executions ?? [], which yields misleading idle states while the query is loading
or errored; update useAgentStatusMap to read isSuccess (from
useGetV1ListAllExecutions) and only run the memoized computation when isSuccess
is true, otherwise return a neutral result (e.g., an empty Map or a Map with a
loading/unknown status) instead of deriving statuses from missing data; ensure
you stop using executions ?? [] and only pass executions into
computeAgentStatus(agent, agentExecs) when isSuccess is true.

[coderabbitai]

⚠️ Potential issue | 🟠 Major

Return a loading state here instead of an all-zero summary.

This hook currently reports 0 for every fleet count until executions load, which is indistinguishable from a genuinely empty fleet and can flash incorrect briefing/filter totals in the UI. useLibraryFleetSummary in this same PR already avoids that by waiting for a successful query.

Based on learnings, prefer checking isSuccess when gating on React Query results because data may be undefined in error/idle states.

🤖 Prompt for AI Agents

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

In `@autogpt_platform/frontend/src/app/`(platform)/library/hooks/useAgentStatus.ts
around lines 144 - 208, The hook useFleetSummary currently builds and returns an
all-zero FleetSummary while the executions query is still loading; update it to
detect the query status from useGetV1ListAllExecutions (use its isSuccess flag)
and return a distinct loading state instead of the zeroed summary. Specifically,
inside useFleetSummary, read isSuccess from useGetV1ListAllExecutions and if
isSuccess is false return an explicit loading sentinel (e.g., null or a
FleetSummary with a loading boolean) so callers can differentiate "loading" from
a genuine empty fleet; only compute the counts inside the useMemo when isSuccess
is true and executions is available. Ensure references to useFleetSummary,
useGetV1ListAllExecutions, and FleetSummary are updated accordingly.

[coderabbitai]

⚠️ Potential issue | 🟠 Major

Scope the status filter UI to the same dataset as the active tab.

fleetSummary is built from the global useLibraryAgents() result, but statusFilter is applied to whatever list the active tab is showing. On the Favorites tab, that means the dropdown counts describe all agents while the list only shows favorites. Either compute a favorites-specific summary, or reset/hide the status filter when activeTab !== "all".

One simple way to avoid the mismatch

   function handleTabChange(tabId: string) {
     setActiveTab(tabId);
     setSelectedFolderId(null);
+    if (tabId !== "all") {
+      setStatusFilter("all");
+    }
   }
...
-          fleetSummary={isAgentBriefingEnabled ? fleetSummary : undefined}
-          briefingAgents={isAgentBriefingEnabled ? agents : undefined}
+          fleetSummary={
+            isAgentBriefingEnabled && activeTab === "all"
+              ? fleetSummary
+              : undefined
+          }
+          briefingAgents={
+            isAgentBriefingEnabled && activeTab === "all" ? agents : undefined
+          }

Also applies to: 58-61

🤖 Prompt for AI Agents

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

In `@autogpt_platform/frontend/src/app/`(platform)/library/page.tsx around lines
24 - 27, The statusFilter dropdown is showing counts derived from fleetSummary
(created from useLibraryAgents()) while the UI list is scoped by activeTab,
causing mismatched counts on non-"all" tabs like Favorites; update the logic so
the summary is computed from the same dataset the active tab uses (e.g., derive
a tabScopedAgents list from useLibraryAgents() filtered by activeTab and pass
that into useLibraryFleetSummary) or, alternatively, clear or hide statusFilter
(via setStatusFilter) when activeTab !== "all" so the counts and filter behavior
stay consistent with the displayed agents.

[github-actions]

This pull request has conflicts with the base branch, please resolve those so we can evaluate the pull request.