Skip to content

Persist M5 helper startup failures before detached worker execution #216

Open

Description

@realmarcin

Read-only review of PR #214 found that run_m5_helper_20261119.py execute() validates runtime, prior canary, output and metadata before entering its exception handler. If the prompt or wrapper changes after detached tmux launch but before execution, validation raises without completion.json or a durable failure record. The parent launch already reported success and the pane may disappear.

Move startup checks into durable failure handling after safely validating the owned output directory. Persist failure stage/type without child output or secrets, preserve the original plan and metadata, and never start the provider after validation failure. Add a regression that mutates the prompt or wrapper between staging and execution, then checks the on-disk failure artifact and absence of provider events. Scientific readiness remains unaffected.

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