Skills

Skills are reusable capability packages that make your AI agent better at specific tasks. When you install a skill, Pi learns how to do something it didn’t know before — whether that’s building on a specific platform, running AI models, or following design best practices.

The format was created by Anthropic and has since become an open standard maintained at agentskills.io. Pi implements this standard directly, so skills you install work across Pi and other tools that support the same format.

How skills work

A skill is a folder containing a SKILL.md file. At minimum that file has a name, a description, and instructions for the agent to follow. Skills can also bundle scripts, reference documentation, templates, and other assets:

my-skill/
├── SKILL.md       # required: name, description, instructions
├── scripts/       # optional: executable code
├── references/    # optional: docs and specs
└── assets/        # optional: templates and data

Pi uses skills in two ways:

• Automatically — when your request matches a skill’s description, Pi loads and applies it without you asking.
• Manually — invoke a skill by name yourself with /skill:name, e.g. /skill:replicate find a good image generation model.

Skills use progressive disclosure to stay efficient: at startup, Pi only loads the name and description of each skill — just enough to know when to use it. The full instructions load on demand.

Where skills live

• Global: ~/.pi/agent/skills/ or ~/.agents/skills/
• Project: .pi/skills/ or .agents/skills/ (discovered from your working directory up to the git repo root, only loaded once the project is trusted)

Prefer .agents/skills/ over .pi/skills/ for anything you’d want to reuse outside Pi — it’s the agent-agnostic convention, and other tools (Claude Code, Codex) look for the same directory name.

If you already have skills installed for another tool, you don’t need to duplicate them. Point Pi at those directories too, in ~/.pi/agent/settings.json:

{
  "skills": ["~/.claude/skills", "~/.codex/skills"]
}

Install with npx skills

The best way to install skills is with the skills CLI. It installs to all your AI tools at once — Pi included — from a single command.

You’ll need Node.js installed to run npx skills. If you don’t have it, ask Pi to help you install it for your operating system.

Install four skills

Run these four commands to install a useful set of global skills:

Cloudflare — skills for building on Workers, AI, KV, R2, D1, and the broader Cloudflare platform:

npx skills add https://github.com/cloudflare/skills

Replicate — skills for discovering, comparing, and running AI models via Replicate’s API:

npx skills add replicate/skills

frontend-design — UI/UX design patterns and front-end best practices from Anthropic:

npx skills add https://github.com/anthropics/skills --skill frontend-design

skill-creator — create, test, and improve your own skills from any workflow you want to make repeatable:

npx skills add https://github.com/anthropics/skills --skill skill-creator

Run each command in your terminal. The CLI will detect Pi and set everything up.

Pick them up with /reload

Skills don’t require a restart. Run /reload after installing to pick up the newly added skills, then verify they loaded by prompting “list installed skills”. Then use the prompt below to continue.

Create your own skills

The skill-creator skill you just installed lets you turn any workflow into a reusable skill. After completing some process and wanting to make it repeatable, just tell Pi:

That worked! Let’s create a global skill for it.

Or if it’s something specific to one project:

That worked! I don’t want to have to repeat myself, so let’s create a project skill for this task.

Find more skills

skills.sh is a directory of available skill collections. Browse it to find skills for the tools and workflows you use.

For the full reference — including validation rules and frontmatter options — see Pi’s own skills documentation and the agentskills.io specification.