One of the most annoying things with AI coding agents is this pattern: “Done.” Then you look closer and it never ran the test, build, or app. I built a small tool to force proof before the agent can finish: agent-done-or-not What it does: - wraps the real verification command - records command + exit code + SHA-256 of the output - blocks the agent from finishing unless the latest check is fresh and passing So instead of: “should be fixed” you get: “here is the passing rec...