pi-multi-skills

Multi-skill invocation extension for pi coding agent.

Reference any installed skill from anywhere in your prompt using $skill_name syntax — works inline, not just at the start of a message like /skill:name.

Apply $code-review and $ui-ux-pro-max to review this UI

Run $karpathy-guidelines on the latest changes,
then $interview-me on the architecture

Install

pi install npm:pi-multi-skills

Or install directly from GitHub:

pi install git:github.com/QuangThai/pi-multi-skills

Reload the session:

/reload

Verify installation:

/skills

Usage

Reference one or more skills inline within any prompt:

Apply $code-review and $ui-ux-pro-max to review this UI

Skills resolve regardless of their position in the message:

Run $karpathy-guidelines on the latest changes, then $interview-me on the architecture

A bare skill reference also works:

$interview-me

Autocomplete

Press Tab after typing $ to browse available skills:

Apply $code- [Tab]
  ↓
┌─ $code-change-verification       ─┐
│ $code-review-and-quality          │
│ $code-simplification              │
└───────────────────────────────────┘

Commands

Command	Description
/skills	List every installed skill with $name syntax
/skills-search <keyword>	Search skills by name or description

How it works

Step	Component	Role
1	Resolver	Reads Pi's loaded /skill:name commands so $skill_name respects Pi trust, settings, package filters, and CLI-loaded skills
2	Parser	Extracts $skill_name references from user input using regex
3	Input event	Intercepts user input, expands each $skill_name into a <skill> XML block — the same format Pi's native /skill:xxx command produces
4	Autocomplete	Registers a $-triggered autocomplete provider so the TUI suggests skills as the user types

Comparison: $skill_name vs /skill:name

Aspect	$skill_name (extension)	/skill:name (native Pi)
Position	Anywhere in prompt	Start of message only
Format	<skill name="..." location="...">	<skill name="..." location="...">
Persistence	✅ In conversation history	✅ In conversation history
Multiple skills	✅ Multiple per message	❌ One per message

Both produce identical <skill> XML blocks that the LLM can read. The $skill_name syntax is simply more ergonomic for inline and multi-skill usage.

Architecture

pi-multi-skills/
├── package.json       Pi package metadata
├── index.ts           Extension entry — input event handler + commands
├── resolver.ts        Skill registry from Pi's loaded skill commands
├── parser.ts          $skill_name regex parsing and replacement
├── tests/             Unit tests (29) + E2E integration tests (14)
└── tsconfig.json      TypeScript strict configuration