I got tired of AI assistants confidently giving me broken code. So I built Checkmate. It’s a test-driven dev tool that forces your AI coding agents (like Cursor) to prove their code works before you hit run.
It’s open-source, fast to set up, and built to keep AI output honest.
How it works:
- You define specs and rules in plain text
- The AI writes code and self-checks against those rules
- Checkmate independently verifies that the evidence matches the logic
- If the AI cheats or misses something, Checkmate rejects it
- The agent has to fix it and retry until it passes
It’s like putting an assert statement around the entire prompting process.
I use it now daily with Cursor. Now my AI-generated code I trust, ships faster, and there’s less back-and-forth.
It’s open-source, fast to set up, and built to keep AI output honest.
How it works: - You define specs and rules in plain text - The AI writes code and self-checks against those rules - Checkmate independently verifies that the evidence matches the logic - If the AI cheats or misses something, Checkmate rejects it - The agent has to fix it and retry until it passes
It’s like putting an assert statement around the entire prompting process.
I use it now daily with Cursor. Now my AI-generated code I trust, ships faster, and there’s less back-and-forth.
There’s a WIKI in the repos https://github.com/richardsondx/checkmate where you can learn more about it
Would love feedback, contributions, and bug reports. It’s early but I think this should be a default part of any AI dev workflow.
0.1.7 coming soon with Checkmate Dashboard. I’ve shared some preview on X.