fix(server): report orphaned managed worktrees

[maria-rcks]

Stacked on #19748. Folders inside the managed worktree directories that no thread row references were invisible to storage cleanup. One host had 43 of them, another a single 6.6 GB one.

Cleanup now reports them as kept worktree entries with their full paths and measured sizes, without deleting their files. Failed and removed entries come first, then orphan kept entries ahead of other kept entries so thread results cannot crowd them out of the 200-entry report. Linked worktrees read "No thread uses this worktree"; other folders read "No thread uses this folder, and it isn't a git worktree". Both root/repository/worktree and direct linked-worktree layouts are scanned. Folders referenced by any thread row (including archived, deleted, and subagent threads), ancestors of referenced paths, project workspace roots (including deleted projects), and symlinks are excluded. Path comparisons use canonical paths and ignore case on Windows. A one-hour mtime grace skips new folders while launches and review-guide checkouts are being prepared. Unreadable folders are skipped, and project-root probe errors preserve the root as an exclusion instead of failing the report. The reference query runs on the read-only SQLite worker.

Orphan sizes are cached in memory across sweeps, with cache entries dropped when their paths stop being orphans. New measurement attempts share a 30-second budget per sweep and keep the existing two-million-entry limit. Sizes cut off by the time budget are reported as unknown and retried on the next sweep; folders that cannot be measured are reported without a size and are not re-walked. The settings card shows full paths for orphan rows and measured sizes for kept entries; thread row titles stay unchanged, and kept bytes do not count toward bytes recovered.

On Blacksmith at b741fbb, storageCleanup.test.ts (137 tests), StorageSettings.test.tsx (4 tests), vp run --filter t3 typecheck, and lint for both touched TypeScript files passed (141 tests total). vp run --filter @t3tools/web typecheck previously passed at fa516f0. Regression coverage includes report crowding, canonical and Windows-normalized ancestor paths, fresh versus old folders, project-root probe errors, and a counter proving cached orphans are not walked again. On a dev server with a seeded database copy, the manual cleanup RPC reported a 125-byte orphan linked worktree and a 6,600,000-byte plain folder ahead of thread entries. It did not report the folder containing a live thread checkout or a fresh review-guide checkout as orphans, and all their files remained on disk (report JSON). The real Storage settings "Delete now" action and expanded "Last run" rows showed the full orphan paths and sizes. Native Windows was not run; case-insensitive matching is covered with injected host paths.

Written by gpt-6.1-sol (Codex) in T3 Code on behalf of Maria.

[macroscopeapp]

Approvability

Verdict: Needs Human Approval

Macroscope's review found this PR not approvable — The server now performs filesystem and database scans during storage-cleanup sweeps and exposes newly reported orphan folders with measured sizes in the settings UI. The implementation is substantial new runtime behavior rather than a small isolated fix, so its operational impact merits human review.

No code changes detected at 983bf01. 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	5.0 KiB	5.0 KiB	+7 B (+0.1%)	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	+7 B (+0.6%)	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	0 B (0.0%)	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	0 B (0.0%)	2.0 KiB	✅
Claude	Live turn WebSocket decoded	21.2 KiB	21.2 KiB	0 B (0.0%)	29.3 KiB	✅
Claude	Live turn messages	2	2	0 (0.0%)	8	✅

Baseline: e733a40 · PR result: 983bf01 · 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.

Dismissing prior approval to re-evaluate 6bf79a0