[P1] Preserve failure reporting after hard stops #895

New Issue

2026-07-24T21:37:56-05:00

jcwalker3 commented

2026-07-24 21:37:56 -05:00

Parent epic: #887
Priority: P1

Problem

When a gate fails, the same closure that stops the mutation also forbids posting the diagnostic handoff. The run therefore ends with the failure recorded nowhere durable, and the next worker — frequently a different LLM client — rediscovers it from scratch. Multiple identical blocked cycles have been burned this way.

Acceptance criteria

Add an append-only report_workflow_failure operation.
It remains available after ordinary mutation gates close, including after stop_required: true.
It cannot alter code, labels, issue state, ownership, or leases — append only, enforced, not conventional.
Reports are sanitized (no credentials, no tokens, no raw env) and idempotent.
Each report is linked to issue, session, and ownership generation.
Repeated identical failures are correlated onto one incident rather than duplicated.
Controller inventory exposes unresolved workflow incidents so the next worker sees them before picking work.

Duplicate verdict

VERIFIED not covered. gitea_capability_stop_terminal_report is read-only and returns only a template after reviewer capability denial (#197); it persists nothing and is inert outside terminal mode. Posting the report still requires create_issue_comment, which is exactly what the closed gate forbids. #712 covers failing closed on conflicting queued commands, not reporting.

Parent epic: #887 Priority: P1 ## Problem When a gate fails, the same closure that stops the mutation also forbids posting the diagnostic handoff. The run therefore ends with the failure recorded nowhere durable, and the next worker — frequently a different LLM client — rediscovers it from scratch. Multiple identical blocked cycles have been burned this way. ## Acceptance criteria - Add an append-only `report_workflow_failure` operation. - It remains available after ordinary mutation gates close, including after `stop_required: true`. - It cannot alter code, labels, issue state, ownership, or leases — append only, enforced, not conventional. - Reports are sanitized (no credentials, no tokens, no raw env) and idempotent. - Each report is linked to issue, session, and ownership generation. - Repeated identical failures are correlated onto one incident rather than duplicated. - Controller inventory exposes unresolved workflow incidents so the next worker sees them before picking work. ## Duplicate verdict VERIFIED not covered. `gitea_capability_stop_terminal_report` is read-only and returns only a *template* after reviewer capability denial (#197); it persists nothing and is inert outside terminal mode. Posting the report still requires `create_issue_comment`, which is exactly what the closed gate forbids. #712 covers failing closed on conflicting queued commands, not reporting.

jcwalker3 added the type:feature status:ready workflow-hardening safety observability labels 2026-07-24 21:37:56 -05:00

Sign in to join this conversation.

1 Participants

Notifications

Due Date

No due date set.

Dependencies

No dependencies set.

Reference: Scaled-Tech-Consulting/Gitea-Tools#895