Docs · Agent Runtime
Agent Skills
Agent Skills are modular capabilities that extend Claude-style agents with tool instructions, metadata, and optional resources.
OpenClaw-compatible model
Skillarc supports AgentSkills-compatible folders where each skill is a directory containing a
SKILL.md with YAML frontmatter and instructions. Bundled skills are loaded first, then local skills can override them by slug.What are Agent Skills?
Agent Skills package focused behaviors that teach the assistant how to use tools safely and effectively. Each skill carries metadata (requirements, tags, tools) plus reusable instructions that are injected only when relevant.
Folder layout
Skillarc reads two roots: agent-skills/bundled and optional agent-skills/local.
tree.txt
agent-skills/
bundled/
web-search-assistant/
SKILL.md
tool-use-safety/
SKILL.md
local/
web-search-assistant/
SKILL.md # overrides bundled skill with same slugFrontmatter schema
The SKILL.md frontmatter declares runtime requirements used by the loader.
SKILL.md
---
name: "Web Search Assistant"
description: "Teach the agent to use web search tools for fresh, verifiable answers."
version: "1.0.0"
tags: ["research", "tool-use", "web"]
tools: ["web_search", "web_fetch"]
required_env: ["OPENAI_API_KEY"]
required_binaries: ["curl"]
required_config_flags: ["allowWebSearch"]
enabled: true
---
When freshness matters, search the web, fetch pages, and cite every source.Runtime loading
Use loadAgentSkills() to merge bundled and local skills, then filter by environment variables, config flags, and binary presence.
agent-skills.ts
import { loadAgentSkills } from "@/lib/agent-skills";
const result = loadAgentSkills({
config: {
flags: { allowWebSearch: true },
disabledSkills: ["experimental-skill"],
},
});
console.log(result.active.map((s) => [s.slug, s.source]));
// [["tool-use-safety", "bundled"], ["web-search-assistant", "local"]]Override and filter flow
- Merge: bundled skills load first, then local skills replace matching slugs.
- Filter: disabled or unmet requirements are skipped with structured reasons.
- Inspect: query
/api/agent-skillsto see active vs skipped skills in your current runtime.
Live Snapshot
Active: 1 · Skipped: 1