measure-ai-proficiency

Measure AI Proficiency

Assess repository context engineering maturity and provide actionable recommendations for improving AI collaboration.

This skill works with Claude Code, GitHub Copilot, Cursor, and OpenAI Codex (via the Agent Skills open standard).

Prerequisites

Install the measure-ai-proficiency tool:

pip install measure-ai-proficiency

Workflow

1. Choose Your Scanning Method

Option A: Scan GitHub Directly (No Cloning Required!)

Scan GitHub repositories without cloning them:

# Scan a single GitHub repository
measure-ai-proficiency --github-repo owner/repo

# Scan entire GitHub organization
measure-ai-proficiency --github-org your-org-name

# Limit number of repos scanned
measure-ai-proficiency --github-org your-org-name --limit 50

# Output to file
measure-ai-proficiency --github-org your-org --format json --output report.json

Requirements: GitHub CLI (gh) authenticated with gh auth login

How it works:

• Uses GitHub API to fetch repository file tree
• Downloads only AI proficiency files (CLAUDE.md, .cursorrules, skills, etc.)
• Scans in temporary directories
• Cleans up automatically
• Much faster than cloning!

Option B: Discover Then Clone (Traditional Method)

For organizations wanting more control, first discover which repos have AI context artifacts:

# Find active repos (commits in last 90 days) with AI context files
./scripts/find-org-repos.sh your-org-name

# JSON output for automation
./scripts/find-org-repos.sh your-org-name --json > repos.json

What you get:

• Total repos in organization
• Active repos (with recent commits)
• Repos with AI context artifacts (CLAUDE.md, AGENTS.md, .cursorrules, etc.)
• Percentage baseline for your org
• List of repos to scan

Requirements: GitHub CLI (gh) and jq

Then clone and scan the identified repos.

Option C: Scan Local Repositories

# Scan current directory
measure-ai-proficiency

# Scan specific repository
measure-ai-proficiency /path/to/repo

# Scan multiple repositories
measure-ai-proficiency /path/to/repo1 /path/to/repo2

# Scan all repos in directory (cloned org)
measure-ai-proficiency --org /path/to/org-repos

2. Run Assessment

Most common commands:

# Local scan
measure-ai-proficiency

# GitHub scan (recommended for orgs)
measure-ai-proficiency --github-org your-org-name

3. Interpret Results

Maturity Levels (aligned with Steve Yegge's 8-stage model):

Level	Name	Yegge Stage	Indicators
1	Zero AI	Stage 1	No AI-specific files (baseline)
2	Basic Instructions	Stage 2	CLAUDE.md, .cursorrules exist
3	Comprehensive Context	Stage 3	Architecture, conventions documented
4	Skills & Automation	Stage 4	Hooks, commands, memory files, skills
5	Multi-Agent Ready	Stage 5	Specialized agents, MCP configs
6	Fleet Infrastructure	Stage 6	Beads, shared context, workflows
7	Agent Fleet	Stage 7	Governance, scheduling, 10+ agents
8	Custom Orchestration	Stage 8	Gas Town, meta-automation, frontier

Score interpretation: File count matters more than percentage. The tool includes hundreds of patterns for comprehensive detection.

Understanding Quality Scoring

Each AI instruction file is scored 0-10 based on quality indicators:

Symbol	Indicator	What It Means	Points
§	Sections	Markdown headers (##)	0-2
⌘	Paths	File/dir paths (/src/)	0-2
$	Commands	CLI in backticks	0-2
!	Constraints	never/avoid/don't	0-2
↻N	Commits	Git history (N commits)	0-2

Commit scoring: Files with 5+ commits get full points (indicates active maintenance). 3-4 commits = 1pt.

Cross-Reference Detection

The tool detects links between your AI instruction files:

• Markdown links: [architecture](ARCHITECTURE.md)
• File mentions: "CONVENTIONS.md" or `TESTING.md`
• Relative paths: ./docs/API.md
• Directory refs: skills/, .claude/commands/

Resolution tracking shows if referenced files exist (helps find broken links).

Bonus points: Up to +10 points from cross-references (5 pts) + quality (5 pts).

Content Validation

The tool validates that your documentation references real files:

• Missing references: Files mentioned in docs that don't exist
• Stale references: References to deleted files (detected via git history)
• Template markers: Uncustomized content (TODO, PLACEHOLDER, etc.)

Validation penalty: Up to -4 points for validation issues.

Skip false positives: If your docs contain example file names (meta-tools, templates), configure skip_validation_patterns in .ai-proficiency.yaml:

skip_validation_patterns:
  - "COMPLIANCE.md"    # Example mentioned in docs
  - ".mcp.json"        # Best practice not yet adopted
  - "examples/*"       # All files under examples/

4. Provide Recommendations

After assessment, offer to create missing high-priority files:

Level 2 gaps: Create CLAUDE.md, .cursorrules, or .github/copilot-instructions.md

Level 3 gaps: Create ARCHITECTURE.md, CONVENTIONS.md, or TESTING.md

Level 4 gaps:

• Create skills directories: .claude/skills/, .github/skills/, or .cursor/skills/
• Add .claude/commands/ for slash commands
• Create MEMORY.md or LEARNINGS.md
• Consider SOUL.md or IDENTITY.md (ClawdBot pattern) for agent personality
• Boris Cherny's key insight: Add verification loops (tests, linters) - this 2-3x quality

Level 5 gaps:

• Create specialized agents in .github/agents/ or .claude/agents/
• Set up .mcp.json at root (Boris Cherny pattern) for team-shared tool configs
• Add agents/HANDOFFS.md for agent coordination

Level 6 gaps: Beads memory system, shared context, workflow pipelines

Level 7 gaps:

• Add GOVERNANCE.md for agent permissions and policies
• Set up convoys/ or molecules/ (Gas Town work decomposition)
• Consider swarm/, wisps/, polecats/ for advanced agent patterns

Level 8 gaps:

• Build custom orchestration in orchestration/
• Consider .gastown/ for Kubernetes-like agent management
• Add protocols: MAIL_PROTOCOL.md, FEDERATION.md, ESCALATION.md, watchdog/

5. Create Missing Files

When creating context files, include:

CLAUDE.md structure:

• Project overview (what it does, who it's for)
• Directory structure and key files
• Important conventions and patterns
• Common tasks and how to perform them
• Things to avoid

ARCHITECTURE.md structure:

• System overview and purpose
• Key components and responsibilities
• Data flow between components
• Important design decisions

CONVENTIONS.md structure:

• Naming conventions
• Code organization patterns
• Error handling approach
• Testing conventions

Quick Reference

Common triggers for this skill:

• "Assess my AI proficiency"
• "How mature is my context engineering?"
• "What context files should I add?"
• "Help me improve for AI coding"
• "Check my CLAUDE.md setup"
• "Am I ready for AI-assisted development?"

Customization

Use the customize-measurement skill for guided configuration:

"Customize measurement for my repo"

Or see the manual guide: https://github.com/pskoett/measuring-ai-proficiency/blob/main/docs/CUSTOMIZATION.md