All guides
Guide · openclaw

Creating skills for OpenClaw

OpenClaw runs an AI agent on your own machine, and it's configured entirely with markdown files — which is exactly what AI Project Scoper hands you. This guide shows how to turn your scaffold into a working OpenClaw skill, step by step.

What a "skill" is in OpenClaw

A skill is a plain markdown file (SKILL.md) that tells the agent how to do one specific kind of task. The YAML frontmatter at the top controls when OpenClaw loads the skill; the markdown body below is the actual instructions the agent follows. No code is required for most skills.

This is the same shape as the scaffold AI Project Scoper hands you — so the project brief and docs you already have are most of the way to a skill.

Where skills live

OpenClaw's gateway auto-discovers skills at startup, so a new folder is picked up next launch.

Minimal SKILL.md

---
name: log-analyzer
description: Use this skill to parse server logs and find error patterns.
requires:
  bins: ["grep", "awk"]
---

# Log analyzer

When the user asks about errors in a log file:
1. Run `grep -i error <file>` to find candidate lines.
2. Group by message with `awk` and count occurrences.
3. Report the top 5 patterns with counts and a one-line cause guess.

From your scaffold to a skill

  1. Take your project's brief / AGENTS.md-style file.
  2. Make a folder ~/.openclaw/skills/<project-name>/ and a SKILL.md inside it.
  3. Put a description in the frontmatter so OpenClaw loads it for the right tasks.
  4. Paste your project instructions into the body; trim to the one job you want the agent to do.
  5. Relaunch OpenClaw — it'll discover the skill automatically.

Tips

Official sources