perf(server): stop loading message bodies for thread summaries

[t3dotgg]

Thread summary refreshes loaded every message body and old approval row. Long threads paid that cost during session and checkpoint updates.

Read the newest user-message timestamp and pending approval count in SQL. Keep plan selection, user-input ordering, and the existing message fast path unchanged.

Part of the performance audit. This PR only changes message timestamp and approval count reads. Full reactor reads and plan history are separate work.

Verified:

• 32 tests passed with vp test run src/persistence/Layers/ProjectionThreadMessages.test.ts src/orchestration/Layers/ProjectionPipeline.test.ts from apps/server.
• Targeted lint passed for all seven changed TypeScript files.
• vp run --filter t3 typecheck passed.
• An in-memory SQLite fixture with 10,000 4KiB messages and 1,000 approvals returned the same summary values. The median for the two source reads fell from 99.5ms to 0.16ms across eight warm samples. This is not whole-command latency.

The tests cover thread isolation, empty history, user-message ordering, pending approvals, and summary refreshes that cannot decode message attachments. No browsers or device tests were run, as requested.

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

---

Note

Low Risk
Scoped read-path change in projection summaries with equivalent aggregate semantics and targeted regression tests; no auth, API, or data-migration surface.

Overview
Thread shell summary refresh no longer loads every projected message or approval row when recomputing latestUserMessageAt and pendingApprovalCount. refreshThreadShellSummary in ProjectionPipeline now calls targeted SQL instead: getLatestUserMessageAt (MAX(created_at) for role = 'user') and countPendingByThreadId (COUNT where status = 'pending'). Proposed-plan listing and user-input lifecycle activities are unchanged.

New repository APIs are wired through the persistence service shapes and SQLite layers, with unit coverage for latest-user timestamp isolation and an expanded projection test that corrupts attachments_json and seeds mixed approval/plan rows to prove session/checkpoint-driven refreshes still succeed without decoding message bodies or attachment JSON.

^{Reviewed by Cursor Bugbot for commit abb7dc4. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Stop loading message bodies for thread shell-summary refreshes in ProjectionPipeline

• Replaces loading all projected messages and pending approvals during summary refresh with two targeted repository queries: latest user-message timestamp and pending-approval count
• Adds latest-user-message query to ProjectionThreadMessages.ts and pending-count query to ProjectionPendingApprovals.ts, both scoped to a single thread
• Proposed plans and user-input lifecycle activities still load as before; the refreshed values write back into the existing thread summary row
• Integration test in ProjectionPipeline.test.ts now verifies refresh correctness with corrupted attachment JSON, mixed approval states, and proposed plans
• Risk: refresh logic in ProjectionPipeline.ts no longer reads message bodies or attachment metadata; any downstream consumer expecting those fields on the loaded summary path will not find them

^{Macroscope summarized abb7dc4.}

[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	13.7 KiB	+171 B (+1.2%)	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	6.5 KiB	6.7 KiB	+173 B (+2.6%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	57.0 KiB	58.5 KiB	+1.5 KiB (+2.6%)	66.4 KiB	✅
Codex	Live turn messages	9	10	+1 (+11.1%)	21	✅
Claude	Total thread wire	13.6 KiB	13.6 KiB	−23 B (−0.2%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.0 KiB	7.0 KiB	−4 B (−0.1%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.6 KiB	6.6 KiB	−19 B (−0.3%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	57.9 KiB	57.9 KiB	0 B (0.0%)	66.4 KiB	✅
Claude	Live turn messages	10	10	0 (0.0%)	21	✅

Baseline: 01f3e50 · PR result: abb7dc4 · 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 abb7dc4

Macroscope's review found this PR approvable — This is a focused performance refactor that preserves thread-summary behavior while avoiding message-body and resolved-approval loading during refreshes. The new aggregate queries and thread-isolation behavior are covered by tests, with no schema, deployment, security, or product-default changes.

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