| name | skill-creator |
| description | Create or update MMClaw skills. Use this when the user wants to add new capabilities, automate a specific tool, or define complex multi-step workflows for the agent. |
| metadata | {"mmclaw":{"emoji":"🛠️","os":["linux","darwin","win32"],"requires":{"bins":[]}}} |
Skill Creator (MMClaw)
Use this skill to autonomously design, write, and deploy new skills to MMClaw. A skill is a directory containing a skill.md file that teaches the agent how to use a specific tool or perform a task.
Trigger phrases: "create a skill", "add a new skill", "make a skill for", "teach you how to use", "update the X skill".
1. Directory & File Structure
Skills MUST follow this exact structure to be detected:
- Root Path:
$MMCLAW_WORKSPACE/skills/
- Skill Directory:
$MMCLAW_WORKSPACE/skills/<skill-name>/
- Main File:
$MMCLAW_WORKSPACE/skills/<skill-name>/skill.md
Naming Rules:
- Directory Name: Use lowercase kebab-case (e.g.,
google-search, docker-manager).
- File Name: Always
skill.md.
2. Header Format (CRITICAL)
The skill.md file MUST start with a YAML-style frontmatter block. The formatting of the metadata line is extremely strict (single-line JSON).
Header Example:
---
name: my-skill-name
description: A concise one-sentence summary of what this skill does for the agent's index.
metadata:
{ "mmclaw": { "emoji": "🚀", "os": ["linux", "darwin", "win32"], "requires": { "bins": ["required-binary"] } } }
---
Field Definitions:
- name: Must match the directory name.
- description: This is what the agent sees in its global "Available Skills" list. Make it clear and action-oriented.
- metadata:
- emoji: A single representative emoji.
- os: List of compatible operating systems (
linux, darwin, win32).
- requires.bins: List of CLI binaries that must be installed for this skill to work.
3. Content Structure
After the header, the skill.md should be a well-structured Markdown document containing:
- Overview: When and why to use the skill.
- Trigger Phrases: Common ways the user might ask for this skill.
- Core Invocation: Example shell commands (use
shell_execute or shell_async).
- Parameters/Args: Explanation of available options.
- Examples: Realistic use cases.
- Troubleshooting/Notes: Edge cases or common errors.
4. Workflow for Creating a Skill
- Research: If the skill is for a CLI tool, run
--help or check docs to understand its syntax.
- Create Directory:
mkdir -p $MMCLAW_WORKSPACE/skills/<name>
- Write
skill.md: Create the file with the correct header and detailed instructions.
- Verify: Ensure the file is saved.
- Reload: Call
shell_execute("ls ~/.mmclaw/skills") or simply wait for the next turn; MMClaw will automatically detect the update and print [*] Skill update detected.
5. Implementation Tips
- Be Specific: Give the LLM exact commands to copy-paste.
- Use Placeholders: Use
<variable> or [optional] for dynamic parts of commands.
- Async vs Sync: Instruct the agent to use
shell_async for long-running processes (servers, watchers) and shell_execute for quick tasks.
- Environment: Always remind the agent to check the OS context before generating commands.