dyad:multi-pr-review

Multi-agent pull request review reduces bias by spawning three independent sub-agents to review changes. Each agent receives files in a randomized order, with one focusing on code health and others on correctness and security.

Consensus voting aggregates code review findings by only reporting issues identified by two or more agents. Reported issues must also meet a minimum threshold of MEDIUM or higher severity to ensure high-quality feedback.

Yes, automated code reviews deduplicate findings against existing PR comments. This prevents duplicate feedback and reduces noise by ensuring only new, consensus-backed issues are surfaced in the summary.

The best way to review large pull requests is distributing files among multiple agents with randomized orders. This approach covers diverse file changes while maintaining focused analysis on correctness, security, and code health.

Yes, you need an anthropic dependency to perform consensus-driven PR analysis. The multi-agent coordination relies on Claude-based reviewers to independently analyze diffs and submit JSON-based issue schemas for aggregation.

Yes, diff analysis supports a configurable agent count, consensus threshold, and extended thinking budgets. This allows you to adjust the depth of the code review and the strictness of issue reporting.