Run a consensus-style multi-agent review workflow that inspects a pull request and returns severity-based findings, prioritized action items, and concrete remediation notes. The skill coordinates multiple agent reviewers to compare perspectives, agree on severity, and produce a consolidated report that can be handed to developers or used to drive automation.
Use this skill when you need a thorough, multi-angle review of code changes in a PR — for security checks, quality gates, or when you want severity-prioritized feedback rather than a single-pass review. Not for high-level planning or conceptual design discussions.
Best with multi-agent or orchestration-capable stacks (agents that support Claude, GPT, Gemini or similar LLMs and can run coordinated reviewer workflows).
Multi-agent PR review skill using consensus voting with file-reordering to reduce position bias. Scripts are well-written Python with async Anthropic API calls, proper argparse CLIs, and prompt injection defenses. All scripts exit with usage/API errors when run without required args (expected). SKILL.md is thin — mostly routing/triggers with no concrete workflow steps or output contracts. Missing references/ directory with review prompt templates.
anthropicScripts are production-quality with good error handling and prompt injection mitigation (XML delimiters around diff content). SKILL.md is the weak point — lacks workflow steps, output contracts, and references to required env vars. The skill would benefit from a richer SKILL.md that guides the agent through the orchestrate→aggregate→post pipeline.
