Skip to content

Report terminal job failures before the overall workflow finishes #426

Description

@rldyourmnd

The publisher currently requires the overall run to be completed. A terminal failed job can therefore remain unreported indefinitely while a different job on the same run is queued. The private canary exposed this delay: repair evidence became publishable only after normal supersession cancelled the remaining run.

Add explicit exact-job failure subjects for unfinished attempts without changing their run status or conclusion. Bind repository/run/attempt/job identities, report only a proven terminal failure, and use a distinct durable marker per failure subject. Later failures and partial snapshots must not erase prior subjects; repeated delivery must not duplicate them. Preserve the existing completed-run interface for current callers and keep agent assignment owner-directed.

The private executor can then track run discovery and job-subject receipts separately, with a tested journal migration and bounded publication. No private identities or runtime evidence belong in this public issue.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions