super-review:run

A diff-scoped PR review uses multi-agent specialist dispatch to analyze only the PR diff, then applies an evidence-quoting false-positive gate to confirm each finding with byte-quote verification before including it in a bounded report.

The best way to reduce false positives in AI code reviews is to apply an evidence-confirmation gate that re-checks quoted findings against the actual code, followed by cross-reviewer collision checks and a meta-verification pass to ensure every issue is twice-confirmed.

Yes, you can use Claude Code to review GitHub pull requests by orchestrating parallel specialist dispatch across correctness, security, design, migration, performance, frontend, observability, and testing domains, outputting a capped actionable report.

Multi-agent PR review pipelines handle framework-specific security checks through stack-aware sub-skill loading that auto-detects technologies like React, Next.js, ORM, or crypto libraries and loads corresponding framework and security catalogs for targeted analysis.

Automated PR review works with both explicit GitHub PR URLs or numbers and automatic diff-from-main detection, ensuring the multi-phase review pipeline always targets the correct set of changed lines for bounded reporting.

AI pull request reviews report issues outside the changed code because they lack a diff-scoped boundary and collision check. Applying a bounded reporting approach with an Opus meta-verification phase keeps findings strictly within the PR diff.