Skip to main content

Chisel AI Tool Use Policy

Summary​

Contributors may use AI tools to assist with their work, but must:

  • Keep a human in the loop - All AI-generated content must be reviewed and understood by the contributor before submission
  • Take full accountability - The contributor is the author and is responsible for their contributions
  • Be transparent - Label contributions containing AI-generated content with a message trailer: Assisted-by: <tool>:<model>, e.g., Assisted-by: Claude Code:claude-sonnet-4-6. Include this trailer in commit messages, Pull Requests, or wherever authorship is normally indicated, regardless of the scope of the contribution.
  • Ensure quality - Contributions should be worth more to the project than the time required to review them

What This Means​

Allowed:

  • Using AI tools to generate code that you review, understand, and can explain
  • Using AI for documentation that you verify for correctness
  • Using AI to help debug or optimize code you understand

Not Allowed:

  • Submitting AI-generated code without thorough human review
  • Using automated agents that take action without human approval (e.g., GitHub @claude agent)
  • Using AI tools to fix "good first issue" labeled issues (these are learning opportunities for newcomers)
  • Passing maintainer feedback to an LLM without understanding and addressing it yourself

Contributors using AI tools must still ensure they have the legal right to contribute code under the Apache-2.0 license. Using AI to regenerate copyrighted material does not remove the copyright.

References​

This policy is closely related to and derived from the CIRCT AI Tool Use policy which is derived from the LLVM AI Tool Use Policy.