refactor(server): detach PR lookup stacks instead of a custom cache

[juliusmarminge]

Follow-up to #13304.

#13304 stopped cached PR lookups from retaining caller thread snapshots by replacing Effect's Cache with a hand-rolled LookupResultCache. The leak comes only from the caller's lazy stack frame, which the Cache lookup fiber inherits. Materializing that frame before Cache.get is enough.

This adds a 20-line detachStackFrame helper, routes the three PR-lookup Cache.get calls through it, and returns to Cache.makeWith. The custom cache and its expiry, eviction, and cancellation tests are removed, since Effect's Cache owns that behavior again.

Verification:

• The isolated GC fixture releases the caller snapshot with the helper for both successful and failed lookups, while keeping the result cached. Bare Cache (original mode) retains it on success; the failed-lookup control retained it in one of two runs.
• detachStackFrame.memory.test.ts and GitManager.test.ts pass (116 tests). Scoped lint, formatting, and the server typecheck pass.

Claude Opus 5.5 via Claude Code in T3 Code.

🤖 Generated with Claude Code

---

[macroscopeapp]

🟡 Medium testing/StackRetention.fixture.mjs:2

On Node 22.16/22.17, the spawned process exits while loading ../detachStackFrame.ts because it is invoked only with --expose-gc; those Node versions still require --experimental-strip-types for .ts imports, so the fixture never reaches its assertions. Add that flag to the child-process invocation or use a JavaScript fixture/built output.

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

In file @apps/server/src/git/testing/StackRetention.fixture.mjs around line 2:

On Node 22.16/22.17, the spawned process exits while loading `../detachStackFrame.ts` because it is invoked only with `--expose-gc`; those Node versions still require `--experimental-strip-types` for `.ts` imports, so the fixture never reaches its assertions. Add that flag to the child-process invocation or use a JavaScript fixture/built output.

[macroscopeapp]

Approvability

Verdict: Would Approve

Macroscope's review found this PR approvable — This is a focused PR-lookup cache refactor that preserves lookup, TTL, eviction, and invalidation behavior while preventing cached fibers from retaining request snapshots. The supplied Medium finding identifies a supported-Node test-fixture compatibility problem, which remains a separate correctness issue.

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: d768a00 · 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: 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.