What problem does it solve? Single-model code reviews carry one model's biases and blind spots. This Skill sends the same review request to several LLMs in parallel, then has the host model verify each finding against the actual code and merge the survivors into one consolidated review. ## Core Features & Use Cases - Parallel multi-LLM review: Dispatches the same prompt to every enabled reviewer in llms.json (including the current Claude session itself) via background processes, with per-model timeouts. - Full-repository context: Checks out the review target into a disposable git worktree so reviewers can compare against existing conventions, search for duplicated utilities, and verify DRY violations instead of judging the diff in isolation. - Guideline-aware prompting: Collects CLAUDE.md, lint configs, and style guides into the prompt so reviewers judge against written project rules. - Meta-review fusion: The host model reviews the reviewers' findings one by one, discarding false positives and duplicates, then writes a single verdict with severity-ranked findings. - Two output modes: Local mode writes review.md to a temp directory; PR mode posts inline comments and a summary to a GitHub pull request via gh api. - Use Case: Before merging a large feature branch, ask for a multi-model review of the local diff; receive one deduplicated, fact-checked findings list instead of four raw model opinions. ## Quick Start Ask the assistant to run a fusion review of the uncommitted local changes and report the consolidated findings.
