[Fix] Pull requests become ready after an explicit draft request

[roomote-roomote]

​Opened on behalf of @daniel-lxs. Follow up by mentioning @roomote-roomote, in the web UI, or in Telegram.

What changed

An explicit draft-state update through Roomote now prevents clean-review automation from marking that PR ready. This includes an already-draft update that makes no remote change. A successful explicit ready update releases the hold; title, description, and base-branch refreshes preserve it.

The hold is persisted on tracked PR associations, applies across subsequent pushes/reviews, and is checked under the same provider-scoped lifecycle lock used by explicit state updates. Provider failures retain the hold, and ready updates must confirm the requested remote state before releasing it. Includes an additive, default-false migration and updated source-control documentation.

Why this change was made

Previously, the clean-review path only knew that Roomote created the PR. It could therefore override an explicit draft request, even immediately after a successful no-op draft update. Persisting that intent preserves the requested review workflow while retaining normal opt-in promotion for unheld PRs.

Impact

Explicit draft intent remains effective across clean reviews, multiple task associations, and new heads. The normal deployment-default draft creation and automatic promotion of unheld PRs remain unchanged. The hold records state updates made through Roomote; direct provider UI/CLI changes do not create a Roomote hold. No merge or approval behavior is added.

How it was tested

• Current-develop baseline (35201fbf): actual update/promotion helpers with real Postgres associations, real Redis, and controlled GitHub protocol reproduced an explicit no-op draft update being promoted to ready.
• Exact pushed head b4380234a376146d271c9abe2857e3e8c9f6dd9e: 133 focused tests passed. Eight real persistence/ordering tests cover no-op intent, default promotion, explicit ready release, provider failure/retry, later associations/new heads, isolation, controlled in-flight ordering, and independent-process lock ownership, durable hold, release, cleanup, and recovery.
• All five provider promotion gates are exercised. Existing transition tests and adverse ignored-ready-response/lock-loss tests pass. External protocol clients are controlled; these are application/persistence tests, not live-provider parity claims.
• Generated migration applied to the existing local development database; all preexisting associations retain the default unheld state. SDK/DB types, pnpm lint:fast, pnpm check-types:fast, pnpm knip, and normal pre-push gates passed. Mintlify validation and broken-link checks passed.
• Adjacent GitHub review-intake/lifecycle suites passed 64 checks at the same head. Full hosted CI passed, including Test, app Docker, backup/restore and upgrade compatibility. Worker-image matrix and Mintlify deployment are expected skips. Exact-head automated review is clean with zero findings or unresolved threads.
• Judgement is explicitly incomplete despite its green policy result: precommit evaluated 37 judgments with generated-snapshot file-budget gaps; hosted final run evaluated zero judgments and reported evidence/judgment failures plus snapshot budget limits. This is not complete semantic coverage or merge certification.

Checklist

• The PR title follows the repo convention: [Fix], [Feat], [Improve], [Refactor], [Docs], or [Chore] followed by a user-facing description
• This PR is small and scoped to one change (the large JSON addition is the generated Drizzle schema snapshot)
• pnpm lint and pnpm check-types pass locally — normal fast repository gates passed; full formatting-inclusive commands were not run
• I added tests or included a clear manual validation note above
• I removed secrets, tokens, private keys, and customer data from code, logs, and screenshots
• If this change should appear in the changelog, I ran pnpm changeset

Screenshots

Genuine local Mintlify desktop rendering, opened and independently reviewed. This screenshot proves documentation copy/layout only; backend behavior is covered by the persistence/protocol tests above. The existing table-of-contents prior-section highlight is visible. No simulated UI state was used.

Related PRs

• #3444 — Restore native Brain helper expansion and synthesis: independent reliability work from the same shift; neither PR depends on the other.
• #3445 — Bound controller shutdown while launches stall: independent controller reliability fix from the same shift.

[roomote-community]

No code issues found. See task

Reviewed draft-intent persistence, explicit state updates and provider confirmations, promotion locking, association refreshes, migration consistency, regression coverage, and documentation. Current-head tests, lint, type checks, Knip, docs validation, Docker build, and CodeQL passed.

_{Reviewed b438023}

[daniel-lxs]

Verified explicit draft holds and release through the actual PR writer and clean-review promotion path, with real persistence, locking and a controlled HTTP provider. Focused SDK/database checks and current CI pass at this head.