What problem does it solve? Teams adopting an agent harness often cannot tell which artifacts (entry instructions, context docs, state files, feedback commands, evaluation rubrics, skills) exist or are missing in a repository. This Skill performs a read-only audit of any project and reports exactly what is present, what is missing, and what to fix first. ## Core Features & Use Cases - Six-layer audit: Inspects Entry (AGENTS.md, CLAUDE.md, README.md), Context (docs/ARCHITECTURE.md, docs/PRODUCT.md, docs/RELIABILITY.md), State (feature_list.json, .specs files, progress docs), Feedback (init scripts, test/build/lint commands), Evaluation (judges, rubrics, CI checks), and Skills (harness-* skills). - Fresh-session test: Answers orientation questions (what the repo does, how to start it, what is unfinished, next task) using only files found in the repo. - Non-mutating guarantee: Produces a status table and a minimal bootstrap plan, then stops and asks for approval before writing anything. - Use Case: Run it on a newly cloned repository before starting agent work to discover that no AGENTS.md exists and the documented startup command fails, then get a prioritized fix plan. ## Quick Start Ask the agent to run a harness-init audit on the current repository and report which harness layers are present or missing.
