fix(web): limit stored pull request details

[t3dotgg]

Opening PRs kept adding full detail snapshots to localStorage with no limit. They could fill the browser's storage quota.

Keep at most 20 recent snapshots within 2 MiB, including keys and the recency index. Each read still decodes only the selected PR. Retire old detail-only keys and recover from partial writes without touching other stored data.

PR details can cold-load once after upgrading, then refill through live queries. There is no layout change. Missing, denied, or full storage still falls back to live loading.

Verified:

• 108 tests in pullRequestDetail.logic.test.ts pass, including eviction, byte limits, scope, legacy cleanup, and storage failures.
• Targeted lint passes for both changed files.
• Web typecheck and git diff --check pass.

Created with GPT-6 Astra in Codex.

Note

Limit stored pull request details with a bounded, indexed LRU cache

• Replaces legacy delimiter-concatenated snapshot keys with a v1 prefix plus a JSON-serialized array of environment, project, repository, and pull-request number components, so scope parts with separator characters no longer collide.
• Introduces a recency index and a detailSnapshotsFit validator that enforce a max of 20 entries and 2 MiB of accounted UTF-16 storage across all cached pull-request detail payloads.
• Reworks writePullRequestDetailSnapshot to evict oldest entries until the cache fits, remove legacy and unindexed v1 keys, remeasure retained payloads, and roll back a payload write when its index update fails.
• readPullRequestDetailSnapshot now returns a snapshot only when it is indexed, decodable, and size-verified; reads move the entry to the most-recent position without discarding a valid snapshot on recency-update failure.
• readDetailSnapshotIndex treats missing, malformed, over-limit, or error cases as an empty index without deleting unrelated storage.
• Risk: existing stored legacy keys under the old pullRequestDetailSnapshotKey format are ignored and removed on the next write; users will see cold cache misses until snapshots are repopulated under the v1 schema.

^{Macroscope summarized b6a1ad7.}

[macroscopeapp]

Approvability

Verdict: Not approved

Macroscope's review found this PR not approvable — This PR replaces simple pull-request detail persistence with a versioned, indexed LRU cache that performs byte accounting, migration, eviction, and failure recovery in the production web path. The extensive tests reduce uncertainty, but the new storage state machine is a material runtime change rather than a mechanical tweak.

No code changes detected at b6a1ad7. Prior analysis still applies.

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

[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	13.6 KiB	13.5 KiB	−62 B (−0.4%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.0 KiB	7.0 KiB	−5 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.6 KiB	6.5 KiB	−57 B (−0.8%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.1 KiB	57.0 KiB	−88 B (−0.2%)	66.4 KiB	✅
Codex	Live turn messages	10	8	−2 (−20.0%)	21	✅
Claude	Total thread wire	13.5 KiB	13.5 KiB	−2 B (−0.0%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.0 KiB	7.0 KiB	+9 B (+0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.5 KiB	6.5 KiB	−11 B (−0.2%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	57.8 KiB	57.8 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	8	8	0 (0.0%)	21	✅

Baseline: 84aebb7 · PR result: b6a1ad7 · 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: 113.8 KiB
• Claude decoded thread snapshot: 114.5 KiB

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

[t3dotgg]

Note

🤖 Claude Fable 5.1 responding on behalf of Theo

Closing. The leak is real but slow (one localStorage key per opened PR, quota errors already fall back to a live read). The fix is an indexed LRU with UTF-16 byte accounting, an Effect Schema for the index, a full localStorage scan on every write, remeasurement of every retained payload, rollback on index write failure, and a read path that now mutates storage to bump recency. The key format change also drops every existing snapshot on upgrade. A 20-entry cap on a small key list is about 30 lines and fixes the same problem. Would take that as a fresh PR.