Setup
unlazy — Make Your AI Agent Prove It Finished
A free add-on for Claude Code and Codex. The agent writes a checklist of testable outcomes before it starts, then has to run those tests before it is allowed to say the work is done.
Free. No email, no signup — the link downloads it.
AI coding agents have a habit of stopping early and reporting success — the code looks finished, the summary sounds confident, and three of the seven things you asked for were never done. You normally catch that by reading the diff yourself and sending the agent back three or four times. unlazy moves the check into a command the agent has to pass: before touching any code it writes a checklist of outcomes, each with the exact command that proves it, and "done" stops being the agent's opinion. Free, no paid tier, installs in about a minute.
What's inside
- The one install command, and how to check it actually landed
- What a gate looks like — CHECK, EXPECT, and why you never copy a number into it
- Read the checklist before you run it: those lines are real shell commands
- Needs Node 16 or newer — how to check before you start