fix(server): preserve manual thread titles during generation

[rynfar]

A delayed first-turn title or provider metadata update can overwrite a manual rename, including when the user renames a thread back to the same text. Persist a command-versioned title intent and accept automatic title completions only while the expected title and version still match. Archive/reopen, deletion, a newer title, and explicit regeneration invalidate stale completions. Imported history remains conservative; older event payloads still decode.

This is a partial adaptation of upstream 7cafe52bb8f1a4f26e9a8504c53ceae5cb32548a. It keeps Pylon's existing title-generation call count. The upstream automatic post-turn refinement would add another model call and remains a separate product/cost decision; provider-link refactors 08abda9dcf6fee89561f861004efb48bad665df2 and a62e7d670c67bf221a5699b5a988367781fead74 are outside this fix.

Depends on #815 at refreshed head 360f814b9161e6af0a9e4ed64e80f445755f52b2, merged into this branch. Merged Relay recovery owns migration 063, #815 owns migration 064, and this PR adds migration 065. For focused title-only review, compare the #815 prerequisite head with this PR head.

Validation: 169 focused contract/server tests, 153 ingestion/snapshot/decider tests, and final 15 targeted title tests passed; contracts, client-runtime, and server typechecks passed. Pinned Effect diagnostics checked 1035 server files with zero diagnostics. Scoped lint/format and diff checks passed. Independent read-only review passed. No live provider or browser session was exercised.

The independently useful part of upstream a62e7d670c67bf221a5699b5a988367781fead74 is adapted here: title-generation and recovery warnings now log only a failure category and reason count, rather than a formatted cause that can include provider CLI output. Its provider link-decoding changes still depend on the separate 08abda9d lookup feature.

Refreshed head 82795d32bcdabcfe7a1d999826f16f31171d07e3: merged the updated #815 prerequisite and current pylon cleanly. Eight focused test files passed (352 tests); server, contracts, and client-runtime typechecks, scoped lint, diff check, and frozen offline install passed. Final-head CI is pending.

CI follow-up: the merged #815 prerequisite exposed three repository-wide lint errors in existing reactor tests; this branch changes those four dispatch calls to the harness-owned Effect runtime. Focused approval/user-input tests, scoped lint and format passed.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
pylon-marketing	Ready	Preview	Sep 26, 2026 6:55am UTC

[vercel]

Deployment failed for project pylon-marketing with the following error:

Resource is limited - try again in 24 hours (more than 100, code: "api-deployments-free-per-day").

Learn More: https://vercel.com/pylon-code?upgradeToPro=build-rate-limit

[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	13.9 KiB	14.0 KiB	+79 B (+0.6%)	15.1 KiB	✅
Codex	Thread snapshot wire	7.2 KiB	7.3 KiB	+52 B (+0.7%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.7 KiB	6.7 KiB	+27 B (+0.4%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	58.0 KiB	58.0 KiB	0 B (0.0%)	66.4 KiB	✅
Codex	Live turn messages	9	9	0 (0.0%)	21	✅
Claude	Total thread wire	14.0 KiB	14.0 KiB	+46 B (+0.3%)	15.1 KiB	✅
Claude	Thread snapshot wire	7.2 KiB	7.3 KiB	+48 B (+0.7%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.8 KiB	6.8 KiB	−2 B (−0.0%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	58.9 KiB	58.9 KiB	−44 B (−0.1%)	66.4 KiB	✅
Claude	Live turn messages	10	9	−1 (−10.0%)	21	✅

Baseline: afd6c11 · PR result: 82795d3 · 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: 115.7 KiB
• Claude decoded thread snapshot: 116.4 KiB

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