ENDING A RUN

EXAMPLES Example 1: Objective success criteria A code-fixing agent says it fixed a bug. The harness reruns the tests that failed earlier. Only when they pass does it mark the run complete. Example 2: A structured failure report A run ends with a report that says the status is failed. The reason reads that the database connection timed out three times. The partial progress says that the database structure was analyzed. The recommended next step says to check the database credentials. Example 3: Handoff to the next agent A research agent finishes and passes a short list of key findings with source links to a writing agent. The writing agent never sees the dozens of failed searches, so its context stays clean and focused. Example 4: Cleanup after failure An agent opened a browser session and created temporary files in a sandbox. The run crashed halfway. The always-run cleanup hook closed the browser, deleted the files, and released the database connection anyway. PRACTICE LAB CONNECTION Define success criteria and a failure-report format for the Resume Review Agent lab across Chapters 65 to 68. Test the prompts below against real run outputs from the lab.