is there a way to add custom review prompt ?

[lmcmz]

Great project by the way, very nice to use. 👍
Similar to claude code, is there a way to add custom review prompt ?
https://github.com/anthropics/claude-code-action/blob/main/examples/claude-auto-review.yml#L24-L37

[bstanga]

You cannot change the full prompt, but you can provide custom guidelines. For example:

      - uses: presubmit/ai-reviewer@latest
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}
          LLM_API_KEY: ${{ secrets.LLM_API_KEY }}
          LLM_MODEL: "claude-3-7-sonnet-20250219"
        with:
          style_guide_rules: |
            # Style Guide Rules
            - Prefer descriptive variable and function names
            - Write clear docstrings for all public functions
            - Avoid deeply nested code (more than 3 levels)
            - Follow PEP8 for Python (or equivalent in other languages)
            - Ensure unit tests are present for all new functionality

Those are used directly in the prompt here

Also see discussion about in #37

[m1rl0k]

What I did was fork this; add an input for say Salesforce:on, and clone prompt.ts to my own specs; which allowed me to make my own custom version....

Can also improve chunking/batching to the llm, increase comments; read file line diffs etc to enhance to context/inline comments...

also you can do triggers, and make parts of the prompt conditional if some of the context differs between prs... If a function review this way, if security changes present, add context to prompt etc. Real easy to get creative. May make a fork of this for the devs to see 😇

[bstanga]

Sounds cool! Please consider creating PR to include some of those changes upstream if you think it's useful for other people too. Contributions are welcome

[lmcmz]

Nice, thank you @bstanga :)