Fix worker terminal-state failure misclassification

[alex-solovyev]

Goal

Prevent worker redispatch/noise when an interrupted headless worker already reached a terminal external state.

Context

Recent worker failure excerpts showed a worker process could be terminated after the PR had already merged and the linked issue had closed. The local runtime then recorded a failure because it lacked a final completion signal, even though GitHub state was already terminal.

Files

• .agents/scripts/headless-runtime-worker.sh
• .agents/scripts/tests/test-headless-runtime-helper.sh

Implementation

• Before recording worker_failed, check confirmed GitHub terminal state:
  • linked issue is CLOSED
  • PR for current branch is merged
  • merged PR is discoverable by linked issue number
• If confirmed, release the dispatch claim as worker_complete and skip fast-fail accounting.

Verification

• bash .agents/scripts/tests/test-headless-runtime-helper.sh
• shellcheck .agents/scripts/headless-runtime-worker.sh .agents/scripts/tests/test-headless-runtime-helper.sh
• bash -n .agents/scripts/headless-runtime-worker.sh .agents/scripts/tests/test-headless-runtime-helper.sh
• git diff --check
• .agents/scripts/linters-local.sh

---

aidevops.sh v3.17.28 plugin for OpenCode v1.15.7 with gpt-5.5 spent 3d 18h and 2,985,576 tokens on this with the user in an interactive session.

[marcusquinn]

Interactive session claimed by @marcusquinn on Marcus-MacBook-Pro.local.
Pulse dispatch blocked via status:in-review + self-assignment.

---

aidevops.sh v3.19.2 plugin for OpenCode v1.15.10 with gpt-5.5 spent 11m and 78,560 tokens on this with the user in an interactive session.