Skip to content

worker: harness stop that makes vinci exit 1 is recorded FAILED, not BLOCKED (precedence order) #27

Description

@thegeorgepu

Observed on endurance night 2026-08-29 (box-1, task msg_c03d40fa, daemon 8ad0106, vinci 0.0.52): state=FAILED exit_code=1 harness_stops=1 harness_stop_reason=Blocked (run a command that needs the internet) — no UI to confirm this in a non-interactive run…

finalState (vinci/worker/run.mjs:432-434) ranks exitCode !== 0 above harnessStops.length > 0. When vinci refuses a permission-gated command non-interactively it exits 1, so the stop-caused exit is recorded FAILED although the cause is the harness stop. FAILED is conservative (never a false success), but the terminal misdescribes the cause and hides the BLOCKED count from aggregation.

Proposed: when harnessStops > 0 and the exit is attributable to the stop (vinci's own refusal exit, no crash), BLOCKED wins; non-zero exit with no stop stays FAILED. Needs a fake-vinci test (exit 1 after writing a stop record) + cp-restore mutation proof.

Source: endurance run 2026-08-29 FINDINGS.md.

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