skills2f

Scaffold a Skills 2.0 factory for either Claude Code or Codex.

What it generates

For claude:

• CLAUDE.md
• agents/
• validate.sh
• publish.sh
• skills/
• .env.example
• hermit.yaml

For codex:

• AGENTS.md
• .codex/agents/
• validate.sh
• publish.sh
• skills/
• .env.example
• hermit.yaml

Usage

npx skills2f init-factory mi-proyecto --target claude
npx skills2f init-factory mi-proyecto --target codex
npx skills2f enable-claude-agent-teams
npx skills2f install-skill enreach

The package creates separate outputs for Claude and Codex. There is no both target.

For Claude Code, you can enable Agent Teams automatically with:

npx skills2f enable-claude-agent-teams

This updates ~/.claude/settings.json and sets:

{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

To install a skill directly from Hermit without clawdhub, set:

export HERMIT_URL=https://hermit.cl.n1mbot.cloud
export HERMIT_TOKEN=tu_token

Then run:

npx skills2f install-skill enreach
npx skills2f install-skill enreach 1.0.3
npx skills2f install-skill enreach 1.0.3 /tmp/enreach

You can still override the registry or token explicitly with --registry and --token.

Generated structure

skills/<skill-name>/
├── SKILL.md
├── agents/
│   └── openai.yaml
├── scripts/
├── references/
└── assets/

Notes

• SKILL.md is required.
• agents/openai.yaml is recommended by default.
• The scaffold includes local validation and publish scripts.
• hermit.yaml is a Docker Swarm template for a Hermit portal.
• If an API-backed skill uses mcp2cli, the generated workflow should declare whether it uses pip install mcp2cli or uvx mcp2cli.

Local development

node bin/skills-2-factory.js init-factory /tmp/example-claude --target claude
node bin/skills-2-factory.js init-factory /tmp/example-codex --target codex

Repository

• Private repo: aitorroma/skills-2-factory-npm