review-swarm

To coordinate a multi-agent code review for a pull request, this Skill dynamically spawns adversarial and domain-specific reviewers that collaborate via SendMessage to identify bugs and improvements. It outputs a single, deduplicated combined report classifying findings as BUGs or SUGGESTIONs.

The best way to run adversarial code reviews on a git diff is using a multi-reviewer swarm that dynamically selects specialized agents based on the domain. The swarm collaborates to converge on findings, producing a final, deduplicated report highlighting bugs and improvements.

Yes, you can dynamically select code reviewers for uncommitted changes. The swarm analyzes the diff and domain to spawn up to four dynamic reviewers alongside adversarial and domain reviewers, collaboratively assessing the code changes for issues.

Multi-agent swarm review works across diverse domains including security, concurrency, architecture, and performance. It dynamically selects and applies relevant domain reviewers to the code changes, ensuring specialized assessment of the diff.

A swarm of non-local agent sessions collaborates to find bugs by using SendMessage to communicate and converge on findings. After collaboration, it collates the results into a final, deduplicated report with clear BUG and SUGGESTION classifications.

A limitation of using dynamic LLM reviewers for software quality is the constrained swarm size, spawning up to four dynamic reviewers plus adversarial and domain reviewers. It focuses on identifying bugs and improvements rather than automatically applying code fixes.