harness-context-layer

What problem does it solve? AI agents often make inconsistent architectural decisions or reinvent existing patterns because they lack a documented understanding of the system. This Skill audits the agent's grasp of architecture, product, and reliability, then builds the missing context documents directly from codebase evidence. ## Core Features & Use Cases - Context Layer Audit: Scores the agent's understanding across architecture, product, and reliability categories with a 3/4 gap threshold before proposing any files. - Evidence-Based Doc Generation: Creates ARCHITECTURE.md, PRODUCT.md, or RELIABILITY.md from actual sources like directory trees, package.json, middleware, and CI/CD files — never from assumptions. - Approval-Gated Writes: Produces a scored audit report and waits for explicit approval before writing, creating only one document per invocation. - Use Case: After a major refactor, an agent keeps proposing patterns that already exist. Run the audit, identify that architecture understanding scores 2/4, and generate ARCHITECTURE.md from the real directory tree and entry points. ## Quick Start Audit my project's context layer and tell me which of ARCHITECTURE.md, PRODUCT.md, or RELIABILITY.md is missing or stale.