Files
canina/infra/hermes-editorial/RECOVERY_UI.md
T
2026-10-24 10:58:18 +00:00

1.8 KiB

Owner recovery control

Command center task detail offers ثبت شکست اجرای متوقف‌شده only when the authenticated detail response supplies editorialRecovery: {runId} and that ID matches the registered RUNNING staging run/commit with no artifacts. The server derives the hint from the actual expired outbox lease, canonical editorial admission and unreconciled 20-cent reservation for an independent reviewer. Lists contain no hint. No lease owner/token/credentials are serialized. The existing failure command remains the atomic authority, including provenance, version and run/lease fencing.

Opening the control refetches detail. The modal requires a nonblank reason of at most 500 characters and explicit acknowledgement. It explains retained unknown 20-cent spend, not cost zero, no QA, no autoapproval, no release/retry/budget authorization. A first submit freezes exact runId/expectedVersion/idempotencyKey/reason bytes; ambiguous network retries reuse them. 4xx rejections require refresh. Successful commands and close reconcile with a fresh list. Background content is inert while the recovery modal is mounted.

This adds a backend read-only detail hint because the old safe run projection contains neither lease expiry nor a trustworthy recovery capability. A frontend-only age heuristic cannot establish actual expiry. Deploy both admin and backend changes through normal CI. No change to the failure endpoint, authorization, budgets or worker transport.

Validation: disposable source/dependency copies, no published ports or live state writes. Scoped admin/API tests and application tsc + Vite build; backend hint/failure/read/HTTP tests plus Prisma generation and tsc build. No push/deployment or live task recovery performed here; the owner must authenticate and click after integration.