perf(client): reduce thread-list update work

[t3dotgg]

The shared thread list keeps one atom active for every listed thread. One shell update makes the registry revisit those dependencies across the whole environment.

Build aggregate lists from the environment data and share weakly cached values with point reads. Reuse project membership maps when their entries and order stay the same. Keep thread identity, list order, project moves, and multi-environment behavior unchanged.

In a 10,000-thread fixture, the list and project-reference groups retain 5 registry nodes instead of 10,008. Median update time fell from 41.2 ms to 3.4 ms. This measures the real reducer and shared atoms, not React rendering or device performance.

All 21 focused tests, client-runtime typecheck, and targeted lint pass. An independent source review found no blocker. No wire changes, browser, device, or live state use.

Part of #9661.

Created with GPT-6 Astra (preview) in Codex.

---

Note

Medium Risk
Changes core client reactivity graph for thread lists; behavior is heavily tested but incorrect identity or ordering would affect all thread UI consumers.

Overview
Thread list updates no longer fan out through one atom per thread. Aggregate threadShellsAtom and threadShellsForProjectRefsAtom now walk environment snapshot threads and build scoped shells directly, instead of subscribing to each threadShellAtom in the list.

Scoped shell objects are shared between list reads and point reads via a WeakMap cache keyed by source OrchestrationThreadShell, so list composition does not keep per-thread shell atoms alive. Per-thread threadShellAtom uses the same cache. environmentThreadRefsByProjectAtom also reuses the previous map when project keys, order, and ref arrays are unchanged.

Tests extend the harness for multiple environments and add coverage for registry node count (~200 threads stays under 20 nodes), reference identity after partial updates, and multi-environment ordering, project moves, removals, and catalog changes.

^{Reviewed by Cursor Bugbot for commit 999fa14. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Reduce thread-list update work in createEnvironmentThreadShellAtoms

• Adds a WeakMap-based cache that scopes each source thread shell per environment and reuses the value across point reads and aggregate lists, eliminating one point atom per listed thread
• Project-reference grouping now reuses the previous outer map when project order and per-project reference arrays are unchanged, preserving object identity across unrelated updates
• Global thread list and project lists now read environment indexes and catalog order directly through the shared cache instead of traversing the thread-reference list and point-atom family
• Adds multi-environment test coverage for identity sharing, list ordering, deduplication, catalog filtering, and environment emptying in entities.test.ts
• Risk: aggregate read ordering and membership now follow current project references, catalog entries, and environment snapshot threads directly — callers relying on stale point-atom values may see different results

^{Macroscope summarized 999fa14.}

[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	13.8 KiB	13.5 KiB	−236 B (−1.7%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.0 KiB	7.0 KiB	−2 B (−0.0%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.5 KiB	−234 B (−3.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.5 KiB	57.0 KiB	−1.5 KiB (−2.6%)	66.4 KiB	✅
Codex	Live turn messages	10	8	−2 (−20.0%)	21	✅
Claude	Total thread wire	13.7 KiB	13.5 KiB	−199 B (−1.4%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.0 KiB	7.0 KiB	+2 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.7 KiB	6.5 KiB	−201 B (−2.9%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	59.3 KiB	57.8 KiB	−1.5 KiB (−2.5%)	66.4 KiB	✅
Claude	Live turn messages	10	9	−1 (−10.0%)	21	✅

Baseline: da7e46d · PR result: 999fa14 · 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.

[macroscopeapp]

Approvability

Verdict: Approved at 999fa14

Macroscope's review found this PR approvable — This is a contained client-runtime performance refactor that reduces per-thread atom work while preserving thread data, scoping, ordering, and update behavior. Targeted tests cover identity reuse, large lists, environment/project changes, and lifecycle updates.

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