fix(server): background PR checks spend less GitHub quota

[juliusmarminge]

Every T3 server re-asks GitHub about every unsettled branch once a minute (gh pr list --head, a GraphQL call). The PR lookup cache TTL was also 60s, so nearly every sweep missed it. The settlement sweep added an uncached re-check for every thread whose linked PR was merged or closed, whether or not the thread would settle. Several servers share one token, so traces showed ~2000–2900 gh calls/hour per machine and repeated GraphQL rate limiting.

Changes

• PR lookups without an open PR are now cached for 5 minutes instead of 1. That covers "no PR yet", merged and closed. Branches with an open PR are still checked every minute. Turn end, push, create PR and a user refresh already skip this cache, so PRs opened in-app appear immediately. A PR opened outside the app shows up within 5 minutes.
• The settlement sweep only runs its uncached branch re-check when the linked PR is merged or closed and the thread group would actually settle.

Tests

• New tests for both behaviours in GitManager.test.ts and ThreadSettlementReactor.test.ts.
• The retains protected worktrees … storage-cleanup tests in ThreadSettlementReactor.test.ts fail on main too. This PR does not change them.

Estimated savings

These are estimates from existing server traces (~20 min per machine). Each machine's current unsettled branches were checked against the open PRs in pingdotgg/t3code.

Model

• Branch lookups: before, every unsettled branch cost about 1 gh pr list per minute. After, branches with an open PR still cost 1 per minute and the rest cost 1 per 5 minutes.
• Settlement re-check: now only runs when a thread is about to settle, so it is ~0 in steady state.
• Unchanged: PullRequestSyncReactor, status refresh and checks.

Machine	Unsettled branches (open PR)	Before (calls/h)	After (calls/h)	Cut
mbp (v2)	32 (7)	~2000	~650	~65%
cups	17 (3)	~2900	~2200	~25%
nucbox-1	1 (1)	~300	~300	~0%
macmini (v2)	—	~60	~60	—
Fleet (one shared token)		~5300	~3200	~40%

Where the savings come from

• mbp: branch lookups fall from ~1250/h to ~470/h, because 25 of 32 branches have no open PR. The settlement re-check falls from ~430/h to ~0.
• cups: about two thirds of its calls are PullRequestSyncReactor (one gh pr view per open linked PR per minute), which this PR does not touch.
• nucbox-1: 90% of its calls are PullRequestSyncReactor, so this PR does nothing for it.

Caveats

• The v2 machines need a port. mbp and macmini run the v2 services, so their share only lands once this is ported to v2. This PR alone saves the fleet ~15%.
• The "before" numbers are low. They include time spent rate-limited and backing off, so real demand was higher. The fleet was at or above the 5000/h GraphQL budget, so ~3200/h leaves real headroom.
• The saving grows with the share of branches that have no open PR.

Next: batch PullRequestSyncReactor into one aliased GraphQL query per repo per sweep. That would take cups from ~1900/h of gh pr view to roughly 100/h, and with the v2 port the fleet would sit under ~1500/h.

Follow-ups: port to the v2 services, and batch the per-PR gh pr view in PullRequestSyncReactor.

Done with Claude Opus 5.5 in Claude Code via T3 Code.

🤖 Generated with Claude Code

---

Summary by CodeRabbit

• Bug Fixes
  • Open pull requests continue to refresh every minute, while merged, closed, and no-open-PR results are cached for five minutes.
  • Improved thread settlement handling for groups of threads sharing a branch. Terminal pull-request results are checked against settlement eligibility before a branch refresh, and groups that do not need settlement skip that refresh.
  • Improved handling of resumed threads with a recent user message and a linked pull request reported as merged.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.5 KiB	13.5 KiB	+27 B (+0.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.1 KiB	7.1 KiB	+4 B (+0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.5 KiB	6.5 KiB	+23 B (+0.3%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	56.3 KiB	56.3 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	10	10	0 (0.0%)	21	✅
Claude	Total thread wire	13.5 KiB	13.5 KiB	−8 B (−0.1%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.1 KiB	7.1 KiB	+13 B (+0.2%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.5 KiB	6.5 KiB	−21 B (−0.3%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	57.1 KiB	57.0 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	10	9	−1 (−10.0%)	21	✅

Baseline: 6975efd · PR result: 5294f96 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 114.0 KiB
• Claude decoded thread snapshot: 114.7 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.

[macroscopeapp]

Approvability

Verdict: Approved at 5294f96

Macroscope's review found this PR approvable — This is a focused server-side optimization that reduces background GitHub requests with a bounded five-minute cache for non-open results and preserves explicit refresh behavior. The settlement guard and cache semantics are covered by targeted tests, with no schema, security, billing, or deployment impact.

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

[coderabbitai]

Navigate logical layers of code changes, visualize relationships, and explore their blast radius.

Important

Review skipped

Review was skipped as selected files did not have any reviewable changes.

⚙️ Run configuration

Configuration used: Repository: pingdotgg/t3code/.coderabbit.yaml

Review profile: CHILL

Plan: Team

Run ID: 1fb1b20a-40b4-4e74-b91f-cdfd1acb0971

📥 Commits

Reviewing files that changed from the base of the PR and between 5294f96 and 5294f96.

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

No actionable comments were generated in the recent review. 🎉

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Repository: pingdotgg/t3code/.coderabbit.yaml

Review profile: CHILL

Plan: Team

Run ID: 6ad64fbc-76e8-4fdb-b6f9-901b453340a3

📥 Commits

Reviewing files that changed from the base of the PR and between 901ebf2 and 5294f96.

📒 Files selected for processing (3)

• apps/server/src/git/GitManager.test.ts
• apps/server/src/orchestration/ThreadSettlementReactor.test.ts
• apps/server/src/orchestration/ThreadSettlementReactor.ts

Included review availability: Your plan provides up to 10 included reviews per hour; 5 remain after this review.

---

📝 Walkthrough

Walkthrough

Branch pull-request caching now uses different durations for open and terminal or absent results. Thread settlement groups check whether settlement can occur before refreshing branch data and reuse normalized results. Tests cover cache expiry and settlement behavior for a resumed thread.

Changes

Pull-request lookup and settlement

Layer / File(s)	Summary
Outcome-based pull-request cache policy apps/server/src/git/GitManager.ts, apps/server/src/git/GitManager.test.ts	Open pull requests use a 60-second cache. Merged, closed, and absent results use a five-minute cache. Clock-controlled tests verify lookup behavior at each interval.
Grouped settlement lookup apps/server/src/orchestration/ThreadSettlementReactor.ts, apps/server/src/orchestration/ThreadSettlementReactor.test.ts	Settlement groups check whether any member would settle before refreshing branch data. The code normalizes terminal results once and reuses pull-request data across group members. A test verifies that a resumed thread with a merged linked-PR summary performs no branch lookup or settlement dispatch.

Priority: ➖ Normal

Estimated code review effort: 3 (Moderate) | ~20 minutes

Change: Bug fix

Suggested reviewers: maria-rcks

Merge Risk: ⚪ Minimal · up to 5294f

No actionable merge-blocking issue is established. Normal checks should confirm the cache timing and settlement tests before merging.

🚥 Pre-merge checks | ✅ 5

✅ Passed checks (5 passed)

Check name	Status	Explanation
Title check	✅ Passed	The title clearly summarizes the main change: reducing GitHub quota use from background PR checks.
Description check	✅ Passed	The description explains what changed and why, and includes tests, impact estimates, caveats, and follow-ups. The template’s checklist section is omitted, but the description is otherwise complete; UI…
Docstring Coverage	✅ Passed	No functions found in the changed files to evaluate docstring coverage. Skipping docstring coverage check. Docstring coverage is scoped to functions touched by this diff. Analyzed 0 functions across 4…
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

✨ Finishing Touches

📝 Generate docstrings

• Commit to this branch
• Create a new PR

🧪 Generate unit tests (beta)

• Commit to this branch
• Create a new PR

---

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

Dismissing prior approval to re-evaluate 5294f96