ENDING A RUN

ARTIFICIAL INTELLIGENCE ASSISTANT PROMPTS [START OF PROMPT] Define machine-checkable success criteria for this task that do not rely on the agent's own self-reported "done" claim: task = [TASK DESCRIPTION, e.g. "match resume to job posting and write report"]. Specify the exact objective check the harness should run to confirm completion. Output format: a success-criteria spec (check name, how to verify, pass/fail condition). [END OF PROMPT] [START OF PROMPT] Given this failed run's data, produce a structured failure report using our standard template (status, reason, partial progress, recommended next step): [PASTE LAB OR PRODUCTION RUN LOG/ERROR OUTPUT HERE]. The report must contain enough detail for a different engineer or a follow-up agent to resume the work without re-reading the raw log. Output format: the filled structured failure report, in JSON. [END OF PROMPT] CHAPTER SUMMARY A run is complete only when objective success criteria are met, not when the agent says so. Outcomes are named as success, failure, or partial success. A structured failure report has status, reason, partial progress, and a recommended next step. Results are handed on in structured form, and cleanup always runs, even after failure.