fix(sessions): Surface failed agent prompt sends to the user

[charlesvien]

Problem

Button, review and skill prompts route through sendPromptToAgent, which fired the send as void (fire and forget). A rejected send (auth failure, sandbox unreachable, agent process died, dropped cloud user_message) was swallowed, so the chat showed "Generated in Xs" with no reply and no error. This is one of the paths behind reports of the agent randomly not responding on local and cloud tasks. The chat-input path already toasted these errors; the button/review/skill paths did not.

Changes

1. AgentPromptSender port returns Promise<void> instead of void
2. sendPromptToAgent catches rejections and shows toast.error with a generic fallback
3. Desktop binding awaits sendPrompt so the rejection reaches the catch
4. Add sendPromptToAgent.test.ts covering reject-with-Error, non-Error fallback and no-toast-on-success

Covered callers: CommentAnnotation, PendingReviewBar, PrCommentThread, useFixWithAgent, SkillButtonsMenu. Both local and cloud sessions throw on a failed send, so both are now surfaced.

How did you test this?

1. vitest run on sendPromptToAgent.test.ts (3/3 pass)
2. vitest run on packages/ui/src/features/sessions (225/225 pass, no regression)
3. biome lint and biome format clean on the changed files
4. Typecheck adds zero new errors (the pre-existing @posthog/quill dist errors are unrelated)

Automatic notifications

• Publish to changelog?
• Alert Sales and Marketing teams?

[github-actions]

React Doctor found no issues in the changed files. 🎉

_{Reviewed by React Doctor for commit e021d7b.}

[greptile-apps]

Prompt To Fix All With AI

Fix the following 1 code review issue. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 1
packages/ui/src/features/sessions/sendPromptToAgent.test.ts:36-58
The two rejection tests share identical structure (mock rejection → call → waitFor toast assertion) and differ only in the thrown value and expected message. Per the team's preference for parameterised tests, they can be collapsed into a single `it.each`, reducing duplication without losing coverage.

```suggestion
  it.each([
    [new Error("Agent server is not reachable"), "Agent server is not reachable"],
    ["boom", "Failed to send your message to the agent. Please try again."],
  ])(
    "surfaces rejection %s as an error toast with message %s",
    async (rejection, expectedMessage) => {
      mockSender.mockRejectedValueOnce(rejection);

      sendPromptToAgent("task-1", "hello");

      await vi.waitFor(() =>
        expect(toast.error).toHaveBeenCalledWith(expectedMessage),
      );
    },
  );
```

_{Reviews (1): Last reviewed commit: "surface failed agent prompt sends to the..." | Re-trigger Greptile}

[github-actions]

No showstoppers. The fix correctly surfaces silently-swallowed send failures via toast, the type contract change is consistent across all usages, and the new tests cover the added error paths. The bot comment is a style suggestion about parameterised tests, not a substantive concern.

New commits pushed (delta classified non_trivial_delta) — stamphog approval dismissed; re-review running automatically.

[github-actions]

Solid bug fix that surfaces silently-swallowed send errors via toast. The type contract change is consistent (one implementation, correctly updated), the catch handler is appropriate, and the new tests already use parameterised form addressing the bot's style suggestion.