Stop Being Your Agent's QA: A Skill That Makes AI Coding Agents Prove Their Work

Every AI coding session culminates in the same scenario: the agent declares "Done ✓," and you waste the next 30 minutes manually testing to confirm the outcome. Fed up with being your agent's test suite, I created stop-manual-testing, a skill that transforms the process: every task concludes with machine-checkable criteria the agent must run and pass before declaring completion.

How it operates Every task concludes with a verification list the agent can execute independently: commands, checks, and expected results. If automation is impossible, the skill guides you through the necessary manual verification and the reasoning behind it. The agent enters a closed-loop convergence, avoiding the "Done ✓" declaration and hoping for the best. stop-manual-testing is modular, allowing parallel check execution, and compatible with various agent-driven workflows.

Kaiji-Z/stop-manual-testing Stop manually testing your AI agent. A skill that establishes a machine-checkable verification system, enabling self-validation and closure within the loop. This skill eliminates over 90% of developers' time spent staring at results and relying on gut feelings. Chinese · English English A skill for coding AI agents (Claude Code / Codex / ZCode / Cursor).

Load it once, and instead of manually navigating the UI and assessing whether the agent improved or worsened in the latest run, the agent develops its own verification system where accuracy is machine-checkable, then iterates inside a closed-loop until convergence. The problem it resolves If you work with AI agents, you probably find yourself spending ~90% of your time manually testing—opening the web page, typing inputs, watching the agent step-by-step, and using your brain to determine if it improved… Try it if you spend more time evaluating your agent's work than creating content—feedback appreciated.