What problem does it solve?
Automated pull request review often generates noisy, vague, non-actionable feedback that lacks severity, line precision, and a clear rationale, slowing down merges and increasing human review load.
Core Features & Use Cases
- Severity-tagged, line-specific findings: Emits up to 5 capped findings per PR using exactly one tag each ([CRITICAL], [ISSUE], [NIT]) with per-line
path/to/file:LINE precision and a one-sentence “why it matters”.
- Noise-reduction via skip rules: Skips drafts, WIP/Draft titles, blocked/no-review labels, bot/aeon authors, duplicate head SHA reviews within the last 2 days, and bot-reviewed-recently cases to avoid piling on.
- Diff-aware review strategy: Reviews full diffs when feasible, otherwise reviews top delta files or falls back to per-file patches, and can early-exit for docs-only/test-only/lockfile-only changes.
- Actionable verdict + consolidated summary: Produces an approval-ready verdict or flags blockers for critical issues, and posts both inline comments and a consolidated review body for downstream parsing.
Quick Start
Ask PR Review to review open pull requests for your configured repositories by running it without setting var, after ensuring memory/watched-repos.md lists the targets.