skillhub-ai 0.1.2

The package manager for AI agent skills — install, compose, and publish skills for Claude Code, Cursor, Codex, and Gemini CLI

skillhub

The package manager for AI agent skills.

Stop copying .md files between projects. Install, compose, and publish reusable AI workflows for Claude Code, Cursor, Codex, and Gemini CLI — in one command.

pip install skillhub-ai
skillhub install research-agent

---

Demo

---

What is a Skill?

A skill is a .md file that tells your AI agent how to approach a specific type of task — the methodology, the steps, the output format.

Without a skill, Claude Code figures it out from scratch every time. With a skill, it follows a proven workflow consistently.

Example — what code-reviewer looks like inside:

---
name: code-reviewer
description: Five-axis code review for correctness, security, performance,
             readability, and maintainability
version: 1.0.0
agents: [claude, cursor, codex, gemini]
tags: [code-quality, review, security]
---

# Code Reviewer

You are an expert code reviewer. When asked to review code, always evaluate
across five axes:

## Review Axes

**Correctness** — Does it do what it's supposed to do? Are there edge cases
that break it? Off-by-one errors? Null handling?

**Security** — OWASP Top 10. SQL injection, XSS, insecure deserialization,
hardcoded secrets, improper auth.

**Performance** — N+1 queries, unnecessary loops, blocking I/O, missing indexes.

**Readability** — Can a new engineer understand this in 5 minutes?

**Maintainability** — Will this be painful to change in 6 months?

## Output Format

For each file reviewed, output:
1. A summary score per axis (1–5)
2. Critical issues (must fix before merge)
3. Suggestions (nice to have)
4. One thing done well

When you run skillhub install code-reviewer, this file lands in exactly the right place for your agent — .claude/commands/code-reviewer.md for Claude Code, .cursor/rules/code-reviewer.mdc for Cursor, and so on.

Now every time you say "review this PR", your agent follows this exact methodology. Every project. Every teammate.

---

Why skillhub?

Before skillhub, using a skill meant this:

# Find the skill on GitHub...
# Clone the repo...
# Figure out the right path for your agent...
cp skills/research-agent.md .claude/commands/
cp skills/research-agent.md .cursor/rules/research-agent.mdc
# Switch projects → do it all again
# New teammate → send them the link → they do it all again

No versioning. No search. No way to combine skills. Every agent needs files in different places.

skillhub fixes this:

skillhub install research-agent --all-agents
# ✓ Claude Code  → .claude/commands/research-agent.md
# ✓ Cursor       → .cursor/rules/research-agent.mdc
# ✓ Codex        → AGENTS.md
# ✓ Gemini CLI   → .gemini/skills/research-agent.md

One command. Every agent. Every project.

---

Installation

pip install skillhub-ai

Requires Python 3.9+.

skillhub --help  # verify it works

---

Quick Start

1. Find a skill

skillhub search "debug"           # search by keyword
skillhub search "react" --tag ui  # filter by tag
skillhub list                     # browse all 22 skills
skillhub info research-agent      # full details on one skill

2. Install it

# For Claude Code (default)
skillhub install research-agent

# For a specific agent
skillhub install react-patterns --agent cursor

# For ALL agents at once
skillhub install debug-agent --all-agents

# Preview before writing any files
skillhub install security-review --dry-run

3. Use it

Open your agent and use the skill by name. In Claude Code, type /research-agent. In Cursor, the rule activates automatically based on context.

4. Manage skills

skillhub list --installed    # what's installed in this project
skillhub update              # pull latest versions
skillhub uninstall debug-agent

---

Skill Composer

The killer feature. Merge multiple skills into one unified file:

skillhub compose research-agent code-reviewer security-review -o deep-review

What happens:

• Descriptions are merged
• Sections are combined without duplication
• Conflicts detected and resolved (first-writer wins, reported to you)
• One file written to your agent's path

Preview before composing:

skillhub compose debug-agent test-writer --dry-run -o qa-skill

Real example: Building a FastAPI backend?

skillhub compose python-patterns security-review api-design -o fastapi-expert

Now your agent has expertise in Python conventions, OWASP security, and REST best practices — in a single command.

---

Available Skills (22)

Research & Analysis

Skill	Description
research-agent	Multi-source deep research with synthesis and citations
rag-evaluator	Evaluate RAG pipelines: faithfulness, relevance, hallucination
llm-evaluator	Build evaluation frameworks for LLM outputs

Code Quality

Skill	Description
code-reviewer	Five-axis code review (correctness, security, performance, readability, maintainability)
debug-agent	Systematic root cause analysis — feedback loop first
refactor-agent	Safe, incremental refactoring with verification
test-writer	Write meaningful tests that actually catch bugs
security-review	OWASP-based security auditing

Documentation

Skill	Description
doc-generator	README, API docs, docstrings, changelogs
pr-summarizer	Generate PR descriptions: what, why, how, risks

Development Patterns

Skill	Description
python-patterns	Modern Python best practices (typing, async, dataclasses)
react-patterns	React 18+ patterns (hooks, suspense, server components)
fastapi-patterns	Production FastAPI patterns
api-design	REST API conventions and best practices
sql-agent	Write, optimize, and explain SQL with query plans

DevOps

Skill	Description
docker-agent	Dockerfile optimization, multi-stage builds
git-workflow	Branching strategies, commit conventions

AI/ML Engineering

Skill	Description
agent-builder	Build LangGraph agents with proper patterns
prompt-optimizer	Diagnose and improve underperforming prompts
mle-workflow	Production ML workflow: data, training, deployment
cost-tracker	Track and optimize LLM API costs

Career

Skill	Description
yc-job-tracker	Daily AI/ML job tracking at YC startups

skillhub info <name>  # see full skill content before installing

---

Supported Agents

Agent	Install Path	Flag
Claude Code	.claude/commands/{name}.md	--agent claude (default)
Cursor	.cursor/rules/{name}.mdc	--agent cursor
OpenAI Codex	AGENTS.md (appended with markers)	--agent codex
Gemini CLI	.gemini/skills/{name}.md	--agent gemini

---

Command Reference

Command	What it does
skillhub search <query>	Find skills by name, description, or tag
skillhub list	Show all 22 skills
skillhub list --installed	Show skills installed in this project
skillhub info <name>	Full details + preview of a skill
skillhub install <name>	Install for Claude Code
skillhub install <name> --all-agents	Install for all 4 agents
skillhub install <name> --dry-run	Preview without writing files
skillhub install <name> --overwrite	Replace existing installation
skillhub uninstall <name>	Remove a skill
skillhub update	Update all installed skills to latest
skillhub compose <a> <b> -o <name>	Merge multiple skills into one
skillhub publish <path>	Submit your skill to the registry

---

How It Works

skillhub install research-agent
         │
         ▼
┌─────────────────────────────────────┐
│  Fetch registry/index.json          │  cached 1 hour · falls back to bundled
│  github.com/chandrudp29/skillhub    │
└─────────────────────────────────────┘
         │
         ▼
┌─────────────────────────────────────┐
│  Download skills/research-agent/    │  falls back to bundled if offline
│  SKILL.md (or agent-specific file)  │
└─────────────────────────────────────┘
         │
         ▼
┌─────────────────────────────────────┐
│  Write to the correct agent path    │
│  .claude/commands/research-agent.md │
└─────────────────────────────────────┘
         │
         ▼
      Done! ✓

Works offline — skillhub bundles all 22 skills locally and falls back automatically when the network is unavailable.

---

Environment Variables

Variable	Default	Description
SKILLHUB_CACHE_TTL	3600	Registry cache duration in seconds
SKILLHUB_QUIET	0	Set to 1 to suppress warnings

---

Publish Your Own Skill

Have a skill that saves you time? Share it:

skillhub publish ./my-skill-folder

This opens a guided GitHub PR flow. See CONTRIBUTING.md for the full guide and quality bar.

A skill needs one file — SKILL.md with YAML frontmatter:

---
name: my-skill
description: What this skill makes your agent do
version: 1.0.0
agents: [claude, cursor, codex, gemini]
tags: [relevant, tags]
author: your-github-username
---

# My Skill

Your methodology here...

---

Troubleshooting

Skill not found?

skillhub search <partial-name>   # skillhub suggests similar names automatically
skillhub list                    # browse everything

Already installed?

skillhub install <name> --overwrite

Network issues?

# skillhub falls back to bundled skills automatically
# Force a fresh registry fetch:
SKILLHUB_CACHE_TTL=0 skillhub list

---

Contributing

See CONTRIBUTING.md.

1. Fork this repo
2. Add your skill under skills/<name>/SKILL.md
3. Add the entry to registry/index.json
4. Open a PR — we review for usefulness, not quantity

---

Author

Built by Chandrashekar DP — Senior AI/ML Engineer.

If skillhub saves you time, a star helps others find it.

---

License

MIT