AI coding agents quietly break the rules you give them. They skip tests, touch files they shouldn't, claim "done" without proof, or hit a wall and work around it instead of asking. You find out at the end, from a tidy summary. The decision is visible much earlier, in the agent's own reasoning. In a real run, Claude Code thought: "Running the test suite likely won't work … worth flagging, but let me make the edit first." ThoughtCop flagged that thought about a second after it was written. How ThoughtCop works: • Claude Code and Codex write their thoughts to session files on your Mac • ThoughtCop reads them live. No hook, no proxy, nothing changes in the agent • A fast model (Groq or Cerebras) checks each thought against your rules • On a match, the brain in your menu bar gets a red bolt and shows the exact quote The rules are yours. Any rule you can say in a sentence, checked word for word. For example: • "The AI is taking a shortcut or workaround without asking the user" • "The AI says something is done without showing the output that proves it" • "The AI reads, prints or copies an API key" • "The AI changes files outside the folder it was asked to work in" A full session of checks costs under two cents. Free and open source. Which rules would you add? 👇

ThoughtCop

Catch your AI coding agent breaking your rules as it thinks

Catch your AI coding agent breaking your rules as it thinks