What problem does it solve? Pull requests often accumulate findings from multiple AI reviewers and CI pipelines, leaving developers to manually deduplicate comments, judge which findings are real, apply fixes, and re-run checks. This Skill manages that entire review lifecycle so a single PR target becomes a structured, cost-aware review cycle. ## Core Features & Use Cases - Multi-tool detection and invocation: Detects CodeRabbit, Codex, Jules, Gemini Code Assist, PR-Agent, Graphite, and Reviewdog via config files, GitHub Apps, or workflows, then invokes them in a cost-aware order using MCP, gh CLI, or REST API. - Deterministic triage and remediation: Normalizes and deduplicates findings, applies consensus scoring, evaluates each finding for accuracy and applicability, and applies or defers fixes according to an autonomy policy (full, approve-high-risk, approve-all). - Compliance, reporting, and learning: Checks changes against claude.md, agents.md, and regulatory domains like GDPR or SOC2, posts a structured PR report, and records per-tool precision data to improve future triage. - Use Case: Point the skill at PR #42 with approve-high-risk autonomy; it collects all reviewer comments, auto-fixes low-risk findings, asks approval for risky ones, batches commits to save CI minutes, and posts a final decision table. ## Quick Start Use the pr-orchestrator skill on PR #42 in owner/repo with approve-high-risk autonomy, auto-detected reviewers, and a final summary posted to the PR.
