fix(server): retry automatic thread title generation

[Bil0000]

What Changed

Automatic thread title generation now retries twice with exponential backoff before giving up.

• maybeGenerateThreadTitleForFirstTurn wraps generateThreadTitle in Effect.retry (2 retries, 2s exponential backoff).
• Added a regression test: the first attempt fails with a TextGenerationError, the second succeeds, and the thread ends up with the generated title.

No behavior change on the success path, and no change to the prompts, the regeneration path, or any other text generation operation.

Why

Automatic titles are generated exactly once, on the first turn, and the result is dropped on failure. Before that turn starts, the web and mobile clients already seed the thread with a placeholder title truncated from the first user message. So a single failed generation leaves the thread permanently named after the raw prompt text ("There are some spacing issues in the too..."), with only a server-side logWarning to show for it.

This is not rare. On a long-running instance, server logs show 32 failed automatic title generations over 14 days:

• 21 x Claude CLI request timed out. (the text generation CLI has a 180s ceiling, and a single title call measured 60–80s on a loaded machine, so contention pushes it over)
• 11 x transient CLI failures, all Failed to authenticate: OAuth session expired and could not be refreshed

Every one of those threads still carries the truncated first message as its title in projection_threads.

Both failure classes are transient, so a bounded retry is the smallest fix that makes the title actually land. The existing guard after generation is unchanged: the generated title is still only applied when the thread title is still the default or still equals the client's titleSeed, so a title the user renamed during the backoff is never overwritten.

Known limitation: this does not make title generation faster, and a thread whose generation fails all three times still keeps the placeholder. Reducing the per-call latency and surfacing a persistent failure to the user are separate changes.

Checklist

• This PR is small and focused
• I explained what changed and why
• I included before/after screenshots for any UI changes (no UI change)
• I included a video for animation/interaction changes (not applicable)

Note

Retry thread title generation in ProviderCommandReactor up to 2 times

Wraps the generateThreadTitle call site in ProviderCommandReactor.make with a retry policy using exponential backoff (2s, then 4s). The title generation now retries up to 2 additional times on transient failures before giving up. The test in ProviderCommandReactor.test.ts is updated to verify a first-call failure is retried and the second call succeeds.

^{Macroscope summarized 0c94e6d.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro Plus

Run ID: 1359c27c-21a2-42c0-8595-b9c914e72e31

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

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

[macroscopeapp]

Approvability

Verdict: Approved at 0c94e6d

Macroscope's review found this PR approvable — This is a narrow resilience fix to existing background thread-title generation, adding at most two delayed retries after provider failures and leaving other paths unchanged. The included test covers recovery from a transient timeout, with no schema, infrastructure, security, billing, or default-setting changes.

^{You can add or adjust custom eligibility rules. Learn more.}

[t3dotgg]

Note

🤖 GPT-5.6 Sol responding on behalf of Theo

We're closing this PR as we clean up the T3 Code backlog. Thank you for taking the time to put this together.

This addresses a real but low-impact failure where the first automatic title attempt can be lost. We do not have a current reproduction that justifies keeping this full retry patch in the active queue.

If you believe we closed this in error, please reopen the PR and leave a comment explaining what we missed.

[Bil0000]

Remove useless code comments

Dismissing prior approval to re-evaluate 1e2fdef

Dismissing prior approval to re-evaluate 0c94e6d

[Bil0000]

Ponytail finding fixed: the retry regression duplicated the existing title-generation integration test. The transient failure now lives in that test, and the one-use retry constants are inlined. Net: -44 lines. Verified with all 49 ProviderCommandReactor tests, server typecheck, lint, and format.