Skill Manager

Unified CLI for managing AI skills across all IDEs.

CRITICAL: Global vs Project Skills

ALWAYS confirm scope before any operation. Default assumptions cause errors.

Scope	Location	When to Use
Global	`~/.ai-skills/` → syncs to `~/.claude/skills/`	Personal skills, shared across all projects
Project	`./.ai-skills/` → syncs to `./.cursor/skills/`	Project-specific skills, version controlled

Mandatory Confirmation

Before install, create, or sync, ALWAYS ask user:

"Should this skill be installed globally (~/.ai-skills/) or for this project only (./.ai-skills/)?"

Default behavior:

If user says "project" or "local" → use ./.ai-skills/ in current working directory
If user says "global" → use ~/.ai-skills/
If unclear → ASK, do not assume

Mandatory Rules

Git tracking: When creating or editing skills, ensure all changes are tracked by Git within the project repository.
Project-first: All skill files MUST be created within the current repository only. Only sync to global (~/.ai-skills/ → ~/.claude/skills/) after testing AND explicit user request. Never auto-sync to global.

Common Mistake to Avoid

WRONG: User says "save to project" but agent syncs to ~/.claude/skills/ RIGHT: User says "save to project" → save to ./.cursor/skills/ or ./.ai-skills/ in CWD

WRONG: Agent creates skill and immediately syncs to ~/.claude/skills/ RIGHT: Agent creates skill in project repo → tests it → user explicitly says "sync to global" → then sync

Skill Guidelines Reference

When creating or modifying skills: references/skill-creator/SKILL.md

Important Paths

Type	Path
Global SSOT	`~/.ai-skills/`
Project SSOT	`./.ai-skills/` (in project root)
Claude Code (global)	`~/.claude/skills/`
Claude Code (project)	`./.claude/skills/`
Cursor (global)	`~/.cursor/skills/`
Cursor (project)	`./.cursor/skills/`

Search Workflow (MUST FOLLOW)

When user asks to search/find skills, follow this fallback chain in order:

Step 1: Check Local Installed Skills

# Check global
ls ~/.ai-skills/
# Check project
ls ./.ai-skills/

If a matching skill is already installed, inform user and stop.

Step 2: Search Custom Sources

Check the Custom Skill Sources table below. Match user's query against the listed categories and skill names.

If user asks about biology, bioinformatics, genomics, drug discovery → search K-Dense-AI/claude-scientific-skills
Browse the repo to find matching skills, read their SKILL.md for details

Step 3: Search skills.sh

npx skills search "<query>"

Or browse https://skills.sh for the leaderboard.

Step 4: Search GitHub

If nothing found above, search GitHub for skill repositories:

Search query: <topic> claude skill SKILL.md
Look for repos with proper skill structure (SKILL.md with frontmatter)

Search Result Format

Always present results to user in this format:

Found: <skill-name>
Source: <skills.sh | Custom: K-Dense-AI | GitHub: owner/repo>
Description: <from SKILL.md description>
Install: <command>

Quick Reference

Task	Command
Search skills	Follow Search Workflow above
Install from skills.sh	`npx skills add <owner>/<skill-name>`
Install from custom source	Clone repo + copy to SSOT
List installed	`ls ~/.ai-skills/` or `ls ./.ai-skills/`
Create new skill	Follow skill-creator guidelines
Sync global→IDEs	Copy from `~/.ai-skills/` to IDE paths
Validate skill	Check SKILL.md has name + description frontmatter

Custom Skill Sources

Beyond skills.sh, these curated repositories contain high-quality skills:

Repository	Focus	Skills Count	Install Command
K-Dense-AI/claude-scientific-skills	Biology, Bioinformatics, Scientific Research	142+	See below

Scientific Skills (Biology Focus)

Repository: https://github.com/K-Dense-AI/claude-scientific-skills

Genomics & Sequencing: alphafold-database, biopython, scanpy, pysam, ensembl-database, cellxgene-census, scvi-tools, pydeseq2

Drug Discovery: rdkit, chembl-database, deepchem, pubchem-database, drugbank-database, diffdock

Clinical/Health: clinical-decision-support, pyhealth, clinicaltrials-database, pydicom, pathml

Lab Automation: opentrons-integration, benchling-integration, lamindb, pylabrobot

Install from this source:

# Clone the scientific skills repo
git clone https://github.com/K-Dense-AI/claude-scientific-skills.git /tmp/scientific-skills

# Install a specific skill (e.g., biopython)
# Global:
cp -r /tmp/scientific-skills/scientific-skills/biopython ~/.ai-skills/
# Project:
cp -r /tmp/scientific-skills/scientific-skills/biopython ./.ai-skills/

# Then sync to IDE
cp -r ~/.ai-skills/biopython ~/.claude/skills/

Installing Skills (via skills.sh)

Search Skills

Browse the leaderboard at https://skills.sh to find community skills.

Install from skills.sh

# Install a skill (goes to current directory by default)
npx skills add <owner>/<skill-name>

# Examples:
npx skills add anthropics/skills/docx
npx skills add vercel-labs/skills/find-skills

Post-Install: Choose Scope

After npx skills add, the skill is downloaded. Then:

For Global: Move to ~/.ai-skills/<skill-name>/
For Project: Move to ./.ai-skills/<skill-name>/

Then sync to appropriate IDE paths.

Manual Install (from GitHub)

# Clone skill repo
git clone <github-url> /tmp/skill-temp

# Copy to appropriate scope
# Global:
cp -r /tmp/skill-temp/<skill-name> ~/.ai-skills/
# Project:
cp -r /tmp/skill-temp/<skill-name> ./.ai-skills/

Sync to IDEs

After installing to SSOT, sync to IDE-specific paths:

# Global sync (from ~/.ai-skills/ to ~/.claude/skills/, ~/.cursor/skills/, etc.)
cp -r ~/.ai-skills/<skill-name> ~/.claude/skills/
cp -r ~/.ai-skills/<skill-name> ~/.cursor/skills/

# Project sync (from ./.ai-skills/ to ./.cursor/skills/)
cp -r ./.ai-skills/<skill-name> ./.cursor/skills/

Creating Skills

Follow references/skill-creator/SKILL.md for:

SKILL.md structure (frontmatter + body)
Progressive disclosure pattern
Resource organization (scripts/, references/, assets/)

Quick Create Template

---
name: my-skill
description: >-
  What the skill does. When to use it (triggers, contexts, examples).
---

# My Skill

Instructions for using the skill...

Scope Flags (Legacy Scripts)

If using legacy Python scripts:

-g, --global: Global scope (~/.ai-skills/)
-l, --local: Project scope (./.ai-skills/)

Validation Checklist

SKILL.md exists with valid YAML frontmatter
name and description fields present
Description includes trigger phrases
No extraneous files (README.md, CHANGELOG.md, etc.)
Resources in correct folders (scripts/, references/, assets/)

Troubleshooting

Issue	Solution
Skill in wrong scope	Delete from wrong location, reinstall to correct path
Sync not working	Verify SSOT path matches intended scope
npx skills fails	Check Node.js installed, try `npm exec skills add`
Skill not triggering	Check description includes usage triggers

skill-manager