Agent loses edge cases
You're using an agent to migrate a Bash deploy script to a more robust Python CLI. Twice it has handed back something that runs the happy path but quietly drops the script's original edge handling — the rollback-on-failure branch and the 'skip if already deployed' guard keep vanishing from its rewrite. How do you diagnose why it keeps losing them, re-steer it, and decide when to just finish it yourself?
missing_behaviors(required: list[str], implemented: list[str]) → list[str][["rollback-on-failure","skip-if-deployed","exit-codes"],["exit-codes=covered","skip-if-deployed"]]out["rollback-on-failure=dropped","skip-if-deployed=untested"][["rollback-on-failure","skip-if-deployed"],["Skip-If-Deployed=COVERED"," rollback-on-failure = covered "]]out[][["deploy","rollback","deploy"],[]]out["deploy=dropped","rollback=dropped"]Treat the AI’s output as a draft to verify, not an answer to trust. Name the specific flaw and the input that triggers it, say how you’d catch it (tests, edge cases, reading critically), and how you’d re-prompt or decompose to get it right.
Vibe & agentic: describe the solution in plain language (or narrate it) and the coach grades your approach.