Debug a Lean proof with AI
Inspect the exact goal at a failing line, test a small change and leave a proof another mathematician can follow.
Debug a Lean proof with AI by starting at a particular line where your argument stops working. Perhaps you know that two expressions are equal, but the lemma you chose requires an inequality as a separate hypothesis. The lean-proof-assistant skill asks the agent to inspect the local context before it suggests another tactic.
For a substantial theorem, supply a short mathematical argument as well as the code. A good next step explains what the new goal should be. You can then check whether Lean agrees before investing time in the rest of the proof.
What it does
- Read the hypotheses and notation that are actually in scope.
- Trace a failed tactic back to the goal it received.
- Find a library result that can replace a handwritten helper.
- Check one proof step at a time in the project's environment.
- Remove temporary inspection commands before the final check.
Install and use
- Give the agent the authorised Lean 4 project and a working toolchain, including its Mathlib dependency when used.
- Ask, “Show me the goal before this line. Which assumption does the lemma need that my argument hasn't supplied?”
- Apply the smallest proposed repair, inspect the remaining goals and check the complete file once the proof closes.
FAQ
Source
Made by Texra (texra-ai). Original skill: lean-proof-assistant, published on GitHub under the MIT licence.
View on GitHub