consensus-loop

What problem does it solve?

Coordinating multiple AI agents to resolve GitHub issues and pull requests requires managing independent solvers, converging their conflicting proposals, and safely executing the resulting plan—all while maintaining strict boundaries between design, implementation, and review. Without a disciplined consensus engine, teams risk inconsistent decisions, scope creep, and broken automation that bypasses essential review gates.

Core Features & Use Cases

• Biased multi-solver consensus: Deploys independent minimal-change, structural, and delete solvers followed by a meta-judge to converge on a single concrete plan, ensuring no single perspective dominates.
• Full lifecycle automation: Manages the complete workflow from GitHub issue intake and design consensus through isolated worktree implementation, multi-role review, and conditional merge or release publication.
• Audit-driven fallback: When no managed work is available, the skill performs repository audits to surface philosophy violations and architecture drift, feeding findings back into the main resolution path.

Quick Start

Use the consensus-loop skill to resolve the attached GitHub issue by running the full unattended consensus engine that designs, implements, reviews, and merges the solution.