@4meta5/skills-cli

CLI for discovering, installing, and managing Claude Code skills.

Installation

npm install -g @4meta5/skills-cli

Quick Start

# Scan your project for recommended skills
skills scan

# Install all high-confidence recommendations
skills scan --all

# List installed skills
skills list

# Add a specific skill
skills add tdd

# Show skill details
skills show tdd

Commands

| Command | Description | |---------|-------------| | scan | Analyze project and recommend skills | | scan --all | Install all HIGH confidence recommendations | | scan --show-alternatives | Show all matches, not just top per category | | add <name> | Install a skill by name | | add <name> from <source> | Install from a specific source | | list | List installed skills | | list --custom | List only custom skills (your own) | | list --upstream | List only upstream skills (from sources) | | list --provenance | Show provenance type for each skill | | show <name> | Display skill details | | remove <name> | Uninstall a skill | | validate | Validate skill files in the project | | update --check | Check for upstream skill updates | | update <name> | Update a skill from its source | | update --all | Update all upstream skills | | hygiene scan | Detect auto-generated test slop | | hygiene clean --confirm | Delete detected slop | | claudemd sync | Sync CLAUDE.md with installed skills | | hook available | List bundled hooks | | hook add <names> | Install hooks to project | | hook list | List installed hooks | | hook remove <names> | Remove hooks from project | | evaluate <prompt> | Test skill routing for a prompt | | embed | Generate skill embeddings | | source list | List configured skill sources | | source add <url> | Add a skill repository | | sync <name> --push | Push a skill to all tracked projects | | stats | Show usage analytics | | init | Initialize skills in current project |

Skill Scanning

The CLI analyzes your project to recommend relevant skills:

skills scan

Output:

HIGH   tdd                    Testing workflow enforcement
HIGH   typescript-strict      TypeScript best practices
MEDIUM security-analysis      Security review for PRs
LOW    aws-cdk                AWS CDK patterns

Confidence levels:

• HIGH: Strong match based on tech stack and file patterns
• MEDIUM: Partial match, likely useful
• LOW: Weak match, may not be relevant

Provenance Tracking

Skills track where they came from:

# See which skills are yours vs upstream
skills list --provenance

# List only skills you created
skills list --custom

# List only skills from external sources
skills list --upstream

Hooks

Hooks make skills activate automatically. Install them once:

# See available hooks
skills hook available

# Install hooks
skills hook add skill-forced-eval semantic-router usage-tracker

# Check what's installed
skills hook list

Available hooks:

| Hook | Purpose | |------|---------| | setup-shims | Prepends project-local shims to PATH (SessionStart) | | skill-forced-eval | Forces Claude to evaluate and activate relevant skills | | semantic-router | Matches prompts to skills using embeddings | | usage-tracker | Logs skill activations to ~/.claude/usage.jsonl |

Skill Sources

Add external skill repositories:

# Add a source
skills source add https://github.com/user/skills-repo

# List configured sources
skills source list

# Install from a specific source
skills add my-skill from my-source

Updating Skills

Keep upstream skills current:

# Check for available updates
skills update --check

# Update a specific skill
skills update tdd

# Update all upstream skills
skills update --all

# Update with security review
skills update tdd --review

Hygiene

Clean auto-generated test artifacts:

# Scan for test-skill-* directories and slop
skills hygiene scan

# Include package subdirectories
skills hygiene scan -r

# Delete detected slop
skills hygiene clean --confirm

CLAUDE.md Sync

Keep your CLAUDE.md in sync with installed skills:

# Sync references to match installed skills
skills claudemd sync

Configuration

The CLI uses these directories:

| Location | Purpose | |----------|---------| | .claude/skills/ | Project-level skills | | ~/.claude/skills/ | User-level skills | | ~/.claude/usage.jsonl | Usage tracking data |

Semantic Routing

Skills activate based on prompt similarity. The router combines:

• Keyword matching (30% weight): Fast regex patterns
• Embedding similarity (70% weight): Semantic understanding

Activation modes:

• IMMEDIATE (>0.85): Force activation via MUST_CALL
• SUGGESTION (0.70-0.85): Recommend via CONSIDER_CALLING
• CHAT (<0.70): No activation

Integration

This CLI integrates with these packages:

• @4meta5/skills - Core library for skill management
• @4meta5/skill-loader - SKILL.md parsing
• @4meta5/project-detector - Tech stack detection
• @4meta5/semantic-matcher - Prompt-to-skill matching
• @4meta5/workflow-enforcer - TDD and workflow enforcement
• @4meta5/chain - Skill chaining and DAG resolution

License

MIT