[t3dotgg]

Another slop audit. A lot of this repo is code that nothing can reach: dead exports, code that only its own tests keep alive, client checks on capability flags that every protocol-2 server has always sent as true, legacy paths for peers that can no longer connect, and fixtures that the October test purge (#22557) left behind.

This PR deletes the parts that need no product decision. Users see no change. It removes about 20.5k lines net across 920 files, and 72 files are gone.

What goes

• Flag checks that can never fail. 26 capability flags are true on all 119 protocol-2 releases, and 2 were never true. Clients checked them about 190 times. The server still sends every flag, because older app-store mobile builds read them. Where a check also waited for the server config, an explicit config check stays. With the checks go the legacy "context as text" send path, single-link PR mode, the client-side command-context fallback, the full-snapshot fallback, and per-update thread continuation.
• API nobody calls. 11 WebSocket methods and 3 HTTP routes that no client calls today and no protocol-2 release ever called.
• Code only tests use. For example the in-memory copy of the SQL projection store (ProjectionStore.layerMemory), the ACP event builders, the Grok background translator, the OpenCode CLI inventory, the sidebar PR snapshot code, and the old mobile connection store.
• Dead code. 23 runtime-event types that nothing sends or reads, ThreadLifecycleService, dead private layers, unreachable timeline branches, dead GitActionsControl modes, props nobody passes, and exports that only their own file uses.
• DesktopBridge. The 40 members marked optional "for older shells" are required now. The desktop window only loads its own bundled UI, and the preload defines every member.
• Purge leftovers. Replay fixtures and recorder scenarios nothing reads, 55 acp-mock-agent flags no test sets, native regression fixtures whose test is gone, and a fixture kit for a closed issue.
• Mobile manual memos. 293 useMemo/useCallback calls in files the React Compiler already compiles. Each removal was checked against the compiled output, so the value still comes from the compiler cache with the same deps. 152 stay, because removing them would drop caching.
• Small stuff. Unreachable APNs remote-start code in the relay, two orphaned Expo 57 patches, 3 deps moved to devDependencies, 8 unused marketing images, dead CSS, debug leftovers, and stale docs.

Review tip: hide whitespace. Many tests only lost an indent level, and the memo removals move function bodies left. With whitespace hidden, the diff is +2,494 / -22,999.

Not in this PR (product calls or refactors): LegacySidebar, legacy Plan mode, OpenCode 1.x, the V1 importer and old data migrations, the 5 flags missing from early previews, the resource monitor, hand-run package scripts, and merging the logic that web and mobile both carry.

One finding for later: vite.config.ts plans to ban memo hooks in mobile once the React Compiler allowlist is empty. The 152 memos kept here show that the ban would drop caching in real cases.

Verification

• vp run check:changed on the rebased branch: format, lint, knip, typecheck of 32 packages, and 4,912 of 4,913 tests pass. The one failure is AcpSessionRuntime.processTree. It fails the same way on main on this machine, because the machine has no /usr/bin/node.
• Each of the 12 work slices passed its own focused tests. An independent review read every added line and every removed condition for behavior changes, and found none.

Created with Claude Opus 5.5 in Claude Code, inside T3 Code. Claude Sonnet 5.5 and Claude Haiku 5.5 helped with the audit. GPT-6.1 Sol in Codex worked on the audit, several slices, and the review.

🤖 Generated with Claude Code

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ The exact PR base did not have a successful artifact. Baseline uses the latest successful main measurement shown below.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	5.0 KiB	5.0 KiB	0 B (0.0%)	6.8 KiB	✅
Codex	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Codex	Live turn WebSocket wire	1.2 KiB	1.2 KiB	0 B (0.0%)	2.0 KiB	✅
Codex	Live turn WebSocket decoded	20.9 KiB	20.9 KiB	0 B (0.0%)	29.3 KiB	✅
Codex	Live turn messages	2	2	0 (0.0%)	8	✅
Claude	Total thread wire	5.0 KiB	5.0 KiB	+31 B (+0.6%)	6.8 KiB	✅
Claude	Thread snapshot wire	3.8 KiB	3.8 KiB	0 B (0.0%)	4.9 KiB	✅
Claude	Live turn WebSocket wire	1.2 KiB	1.2 KiB	+31 B (+2.6%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.2 KiB	21.2 KiB	+41 B (+0.2%)	29.3 KiB	✅
Claude	Live turn messages	1	2	+1 (+100.0%)	8	✅

Baseline: 49597c7 · PR result: abb6f73 · 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: 108.5 KiB
• Claude decoded thread snapshot: 108.8 KiB

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

[macroscopeapp]

Approvability

Verdict: Needs Human Approval

Macroscope's review found this PR requires human approval — This is a repository-wide mixed cleanup that also changes production behavior across authentication, RPC contracts, mobile/web capability gating, desktop IPC, orchestration, and notification delivery. The size, cross-owner scope, and omitted diff content make the runtime impact too broad for automatic approval.

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