Verify Before Shipping

What it does

This skill eliminates the #1 failure mode in AI-assisted development: "hallucinated success." It transforms the definition of 'done' from a verbal claim to a verified report. By integrating a lightweight evidence-capture utility, it forces your agent to run actual terminal commands, capture exit codes, and present raw logs before reporting a task as complete.

How it works

The skill provides a specialized proof.py utility (Python stdlib only) that gates progress. Instead of just running commands, the agent executes them through this wrapper. The wrapper produces a structured evidence block including the command string, the status code, and the specific stdout/stderr logs. If any check fails, the skill prevents the agent from claiming success.

Why use this skill

• Eliminate Guesses: No more "I've fixed the bug" followed by a broken build. You get the receipts.
• CI/CD Readiness: Since it exits with non-zero codes on failure, it acts as a local pre-flight check before you even push to a branch.
• Reproducibility: Every fix includes the exact reproduction command and the passing test run in a single output block.
• Professional Handoffs: Ideal for developers who want their AI agents to deliver work that is already verified against ruff, pytest, or curl-based smoke tests.