audit-claude-md
CLAUDE.md Audit
Analyze and score a project's CLAUDE.md against research-informed best practices (ETH Zurich ICML 2026 + production AI-assisted development workflows). Emphasizes what agents CANNOT discover on their own: behavioral constraints, non-standard tools, and project-specific footguns.
Arguments: $ARGUMENTS
Instructions
Audit the CLAUDE.md (or AGENTS.md / .cursorrules) in the current project and score it against the rubric in references/scoring-rubric.md.
If a path is provided in $ARGUMENTS, audit that file instead.
Phase 1: Discover the AI Context Files
Search for all AI context files in the project:
- Check project root for:
CLAUDE.md,AGENTS.md,.cursorrules - Check for nested CLAUDE.md files in subdirectories (e.g.,
src/CLAUDE.md,docs/CLAUDE.md) - Check for linked guide directories:
docs/claude/,docs/ai/,.claude/
More from skinnyandbald/fish-skills
simplify-parallel
Run code simplification across entire codebase using parallel agents with automatic segmentation and coordination
38critic-review
Unified plan review — stack detection, Context7 staleness scan, multi-model counselors dispatch, and prioritized triage. Three modes: full pipeline (default), --dry-run (copyable prompt), --feedback (analyze external input).
36pr-resolution
Resolve PR review comments. Use when asked to fix PR feedback, given a PR URL, or on '/pr-resolution'.
34ceo-briefing
Research any topic and produce a structured executive briefing optimized for rapid CEO decision-making.
32counselors
Fan out a prompt to multiple AI coding agents in parallel and synthesize their responses.
32capture-learning
Capture comprehensive problem-solving narratives from work sessions, documenting the journey of discovery
32