What problem does it solve? AI coding agents rely on instruction files like CLAUDE.md, AGENTS.md, and .cursorrules, plus Claude Code settings, but these files drift, conflict, leak secrets, or go stale. This Skill audits them against an opinionated baseline so agents get the context they need and settings stay safe. ## Core Features & Use Cases - Four-layer audit: Checks project context coverage, operational guidance and conventions, Claude Code settings hygiene (permissions, secrets, hooks, MCP servers), and multi-agent consistency and drift. - Diagnostic snapshot: Inventories every detected agentic instruction file with line counts, last-modified dates, and tool family, plus settings.json shape and hook counts. - Static and read-only: Never writes to the codebase, runs commands, or makes network requests; optionally generates a descriptive implementation plan after findings. - Use Case: A team adopting Claude Code, Cursor, and Copilot simultaneously runs the audit to discover that CLAUDE.md and .cursorrules give conflicting commit conventions, settings.json contains a broad Bash(*) permission, and documented commands reference scripts that no longer exist in package.json. ## Quick Start Run /agentic-audit in the project root to audit the agentic instruction files and Claude Code settings, then review the findings report written to .architect-audits/agentic-audit/findings.md.
