56 lines
1.7 KiB
Markdown
56 lines
1.7 KiB
Markdown
# Raven's Verdict - Deep PR Review Tool
|
|
|
|
Adversarial code review for GitHub PRs. Works with any LLM agent.
|
|
|
|
> **Status: Experimental.** We're still figuring out how to use this effectively. Expect the workflow to evolve.
|
|
|
|
## How It Works
|
|
|
|
Point your agent at `review-pr.md` and ask it to review a specific PR:
|
|
|
|
> "Read tools/maintainer/review-pr.md and apply it to PR #123"
|
|
|
|
The tool will:
|
|
|
|
1. Check out the PR branch locally
|
|
2. Run an adversarial review (find at least 5 issues)
|
|
3. Transform findings into professional tone
|
|
4. Preview the review and ask before posting
|
|
|
|
See `review-pr.md` for full prompt structure, severity ratings, and sandboxing rules.
|
|
|
|
## When to Use
|
|
|
|
**Good candidates:**
|
|
|
|
- PRs with meaningful logic changes
|
|
- Refactors touching multiple files
|
|
- New features or architectural changes
|
|
|
|
**Skip it for:**
|
|
|
|
- Trivial PRs (typo fixes, version bumps, single-line changes)
|
|
- PRs you've already reviewed manually
|
|
- PRs where you haven't agreed on the approach yet — fix the direction before the implementation
|
|
|
|
## Workflow Tips
|
|
|
|
**Always review before posting.** The preview step exists for a reason:
|
|
|
|
- **[y] Yes** — Post as-is (only if you're confident)
|
|
- **[e] Edit** — Modify findings before posting
|
|
- **[s] Save only** — Write to file, don't post
|
|
|
|
The save option is useful when you want to:
|
|
|
|
- Hand-edit the review before posting
|
|
- Use the findings as input for a second opinion ("Hey Claude, here's what Raven found — what do you think?")
|
|
- Cherry-pick specific findings
|
|
|
|
**Trust but verify.** LLM reviews can miss context or flag non-issues. Skim the findings before they hit the PR.
|
|
|
|
## Prerequisites
|
|
|
|
- `gh` CLI installed and authenticated (`gh auth status`)
|
|
- Any LLM agent capable of running bash commands
|