PR Custody

docs: improve review skill and BUGBOT rules

@AgentWrapperchecks n/achecks…docs/improve-review-skill-and-bugbot → next2 files · +100 −2updated 3mo ago

▸Description

Why

The review skill (agent/skills/review-github-pr.md) only covered the mechanics of posting PR comments but gave no guidance on what to look for when reviewing code. The BUGBOT config (.cursor/BUGBOT.md) had a single rule. Both needed to encode the project's actual conventions so automated reviewers catch real bugs and skip false positives.

What

Review skill: Added "What to Look For" section covering logging anti-patterns, error handling, type annotations, security, and code style — prioritized by severity
Review skill: Added "Project-Specific False Positives" table (9 patterns) so reviewers don't flag intentional conventions like logger.exception() outside except blocks or relaxed naming rules
BUGBOT.md: Expanded from 1 rule to comprehensive coverage — split into "False Positives" (don't flag) and "Actual Bugs to Flag" (f-string logger calls, stdlib logging ban, swallowed exceptions, missing error context, partial type issues, unchecked None access, hardcoded secrets)
Both files are consistent with each other and grounded in ruff.toml, AGENTS.md, and established codebase patterns

How to Test

Review the diff — all content is derived from existing project config (ruff.toml rules, AGENTS.md guidelines, Makefile tooling)
Verify false positive patterns match the actual ruff ignore and per-file-ignores settings
Verify "bugs to flag" patterns match known issues in the codebase (f-string logging is the most common)

Pre-Review Checklist

I have self-reviewed this PR

Notes

Markdown-only changes, no code impact
The two files share the same conventions but serve different consumers: the review skill is for Claude agents doing PR reviews, BUGBOT.md is for Cursor's automated reviewer

🤖 Generated with Claude Code

loading diff…