Reasoning Commons

This is a failure intelligence layer that helps AI coding agents stop retrying the same broken fixes. It captures execution lineage, traces retry chains, and stores root cause memory so agents don't waste tokens debugging identical environment issues across sessions. The 11 MCP tools let agents self-report failures, query known failure patterns, and retrieve suggested fixes for common problems like node-gyp Python version mismatches or Docker cache loops. You'd reach for this when your agent is stuck in a 40-minute debugging spiral on tooling failures it should have remembered from last week. Runs as a local Node.js server with optional PostgreSQL backing and exposes both REST endpoints and MCP tools via streamable HTTP.