elizaos--eliza/packages/skills

@elizaos/skills

Bundled skills and skill loading utilities for elizaOS agents.

Overview

Skills are markdown files that provide specialized instructions for AI agents to perform specific tasks. Each skill contains:

• Frontmatter: YAML metadata including name, description, and configuration
• Body: Detailed instructions, examples, and references

Installation

npm install @elizaos/skills
# or
bun add @elizaos/skills

Usage

Get Bundled Skills Path

import { getSkillsDir } from "@elizaos/skills";

const skillsPath = getSkillsDir();
// Returns absolute path to bundled skills directory

Load Skills

import { loadSkills, loadSkillsFromDir } from "@elizaos/skills";

// Load from all default locations (bundled + managed + project)
const { skills, diagnostics } = loadSkills();

// Load from a specific directory
const result = loadSkillsFromDir({
  dir: "/path/to/skills",
  source: "custom",
});

Format for LLM Prompt

import { formatSkillsForPrompt } from "@elizaos/skills";

const prompt = formatSkillsForPrompt(skills);
// Returns a compact structured skills section for system prompt

Build Command Specs

import { loadSkillEntries, buildSkillCommandSpecs } from "@elizaos/skills";

const entries = loadSkillEntries();
const commands = buildSkillCommandSpecs(entries);
// Returns array of command specs for chat interfaces

Bundled documentation skills

Alongside community-oriented skills, this package ships elizaos, eliza-cloud, and eliza-app-development — concise references for elizaOS runtime concepts, Eliza Cloud as a backend, and building elizaOS-based applications (including the Eliza app repository layout).

Skill Discovery

Skills are loaded from multiple locations in precedence order (later overrides earlier):

1. Bundled skills - Included in this package (skills/)
2. Managed skills - User-installed skills (<stateDir>/skills/)
3. Curated/active - Human- or agent-promoted skills (<stateDir>/skills/curated/active/)
4. Project skills - Project-local skills (<cwd>/.elizaos/skills/)
5. Explicit paths - Via skillPaths option

Skill Format

Skills are markdown files with YAML frontmatter:

---
name: my-skill
description: A brief description of what this skill does
primary-env: node
required-bins:
  - node
  - npm
---

# My Skill

Detailed instructions for the AI agent...

Frontmatter Fields

Field	Type	Description
name	string	Skill name (must match directory name)
description	string	Human-readable description (required)
disable-model-invocation	boolean	If true, skill won't appear in prompts
user-invocable	boolean	If false, can't be invoked via commands
primary-env	string	Primary runtime (node, python, etc.)
required-os	string[]	Required operating systems
required-bins	string[]	Required binaries in PATH
required-env	string[]	Required environment variables
command-dispatch	string	Set to tool to enable tool-dispatch in buildSkillCommandSpecs
command-tool	string	Tool name used when command-dispatch: tool

API Reference

Types

• Skill - Loaded skill with metadata
• SkillFrontmatter - Parsed frontmatter
• SkillEntry - Full skill entry with all metadata
• SkillMetadata - Resolved metadata (env, bins, os)
• SkillInvocationPolicy - Resolved invocation flags
• SkillCommandSpec - Command spec for chat interfaces
• SkillProvenance - Provenance tracking for the learning loop
• LoadSkillsResult - Result from loading skills

Discovery / Resolution

• getSkillsDir() - Get bundled skills path
• clearSkillsDirCache() - Reset bundled-dir resolution cache
• getCuratedActiveDir() - <stateDir>/skills/curated/active/
• getProposedSkillsDir() - <stateDir>/skills/curated/proposed/
• promoteSkill(name) - Move a proposed skill to active atomically

Loading

• loadSkills(options?) - Load from all locations
• loadSkillsFromDir(options) - Load from single directory
• loadSkillEntries(options?) - Load with full metadata parsed into SkillEntry[]

Formatting

• formatSkillsForPrompt(skills) - Format Skill[] for system prompt
• formatSkillEntriesForPrompt(entries) - Same but from SkillEntry[] (respects invocation policy)
• formatSkillSummary(skill) - Single "name: description" string
• formatSkillsList(skills) - Newline-joined list
• buildSkillCommandSpecs(entries, reservedNames?) - Build command specs

Frontmatter

• parseFrontmatter(content) - Parse YAML frontmatter from markdown string
• stripFrontmatter(content) - Return body without frontmatter
• serializeSkillFile(frontmatter, body) - Re-serialize frontmatter + body (learning loop)
• resolveSkillMetadata(frontmatter) - Resolve SkillMetadata from frontmatter
• resolveSkillInvocationPolicy(frontmatter) - Resolve SkillInvocationPolicy
• resolveSkillProvenance(frontmatter) - Resolve SkillProvenance (returns undefined if missing/malformed)