Make any AI (ChatGPT, Claude, Claude Code, Claude Cowork, Codex, any of them) prove its work is finished before it tells you it's done.
By Gina Wang · Kulanet AI.
| File | What it's for |
|---|---|
index.html |
The friendly web page. Start here. One-click "copy the coach" button, a plain-English explanation of why this matters, and an FAQ. (Hosted on GitHub Pages.) |
dod-coach.md |
The whole coach in one file. Paste it into any AI, type start here, answer a few questions, and it builds your personal Definition of Done. |
dod-reference.html |
The full framework as a styled page, for when you want to go deeper: the root-cause / review / verified / adversarial / evidence steps spelled out. |
dod-reference.md |
The same full framework as a plain Markdown file you can download. |
- Open the web page or the
dod-coach.mdfile. - Copy the coach text (or download
dod-coach.mdand attach it). - Paste it into a fresh chat with your AI and type
start here. - Answer the questions. Keep the checklist it gives you.
No coding, no setup, no account. Works in any AI chat.
AI says "Done!" when the work is wrong, incomplete, or broke something else. A Definition of Done makes it prove the work first, like a receipt. Start with one line and you're already ahead:
"Before you tell me you're done, show me one specific thing I can check to verify it actually worked."