fix(server): v2 background GitHub checks spend less quota

[juliusmarminge]

Ports the GitHub quota stack (#13189, #13198, #13200) onto the v2 orchestrator branch.

Problem

v2 has the same background GitHub loops as main, so it drains the shared 5000/h GraphQL quota the same way:

• branch PR lookups re-ask every minute for branches that have no open PR;
• the settlement sweep does an uncached branch recheck for every terminal link, even when the group would not settle;
• PR sync reads summaries one gh pr view at a time;
• fork lookups probe owner:branch heads, which gh pr list never matches.

Fix

• GitManager: answers without an open PR are cached for 5 minutes instead of 1. This cherry-picked cleanly.
• orchestration-v2/ThreadSettlementService: replaces main's ThreadSettlementReactor, so it gets a hand port of the wouldSettle gate. The uncached branch recheck runs only when the linked PR is terminal and the group would actually settle. A group that becomes eligible only after that check waits for the next sweep instead of settling on an unverified link. The zero-lookup test is ported too.
• GitHubPullRequestCli: summary reads are batched into one aliased GraphQL query per host and credential, with a per-PR gh pr view fallback. PullRequestSyncReactor's sweep goes to concurrency 25. v2's ThreadPullRequestService reads through the same summary path at concurrency 8, so it batches as well. This cherry-picked cleanly.
• GitManager: GitHub head lookups skip owner:/remote: selectors. This cherry-picked cleanly.

Estimates are in the main PRs. The per-call savings are the same here.

Verification

• vp test run over ThreadSettlementService, ThreadPullRequestService, GitManager, GitHubPullRequestCli, gitHubPullRequestJson and PullRequestSyncReactor: 362 passed.
• apps/server typecheck is clean.

Made by Claude Opus 5.5 in Claude Code.

🤖 Generated with Claude Code

---

[macroscopeapp]

🟠 High git/GitManager.ts:1633

findOpenPr can return null for an existing PR when more than 10 open PRs share the branch name across forks: limit: OPEN_PR_PROBE_LIMIT truncates the provider response before matchesBranchHeadContext filters it to the checkout's fork. runPrStep then creates a duplicate instead of returning opened_existing. Paginate the provider results or query with an owner-qualified selector before applying the local filter.

🚀 Reply "fix it for me" or copy this AI Prompt for your agent:

In file @apps/server/src/git/GitManager.ts around line 1633:

`findOpenPr` can return `null` for an existing PR when more than 10 open PRs share the branch name across forks: `limit: OPEN_PR_PROBE_LIMIT` truncates the provider response before `matchesBranchHeadContext` filters it to the checkout's fork. `runPrStep` then creates a duplicate instead of returning `opened_existing`. Paginate the provider results or query with an owner-qualified selector before applying the local filter.

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — The PR substantially changes production GitHub polling, caching, batching, and settlement behavior, including a five-minute default cache for terminal/no-PR results. It also adds a static-analysis suppression and leaves an unresolved high-severity risk where the ten-result probe limit can miss an existing PR and create a duplicate.

Not approved because:

• 1 blocking correctness issue found at or above your repo's Minimum Blocking Severity

^{Adjust the Minimum Blocking Severity for this repo — including turning it Off — in Settings. You can add or adjust custom eligibility rules. Learn more.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ No successful main baseline artifact is available yet. This run establishes the initial measurement.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	—	4.9 KiB	—	6.8 KiB	✅
Codex	Thread snapshot wire	—	3.7 KiB	—	4.9 KiB	✅
Codex	Live turn WebSocket wire	—	1.2 KiB	—	2.0 KiB	✅
Codex	Live turn WebSocket decoded	—	20.4 KiB	—	29.3 KiB	✅
Codex	Live turn messages	—	2	—	8	✅
Claude	Total thread wire	—	4.9 KiB	—	6.8 KiB	✅
Claude	Thread snapshot wire	—	3.7 KiB	—	4.9 KiB	✅
Claude	Live turn WebSocket wire	—	1.2 KiB	—	2.0 KiB	✅
Claude	Live turn WebSocket decoded	—	20.8 KiB	—	29.3 KiB	✅
Claude	Live turn messages	—	2	—	8	✅

Baseline: unavailable · PR result: e18ee51 · Source CI: failure

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: 106.1 KiB
• Claude decoded thread snapshot: 106.4 KiB

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