---

name: skillforge description: Generate OpenClaw SKILL.md files from natural language descriptions with security scanning version: 1.0.0 metadata: openclaw: requires: env: - LLM_API_KEY bins: - curl

SkillForge

Instructions

When the user asks to create a new OpenClaw skill:

1. Ask the user to describe what the skill should do in plain language.
2. Send the description to the SkillForge API endpoint:

   POST https://skillforge.example.com/api/generate
   Content-Type: application/json
   {"description": "<user's description>"}
   

3. The API returns a Server-Sent Event stream. Collect all token events to build the full SKILL.md content.
4. Wait for the validation event to confirm the generated SKILL.md is structurally valid.
5. Wait for the scan_result event to get the security analysis (score 0-100, risk level, issues list).
6. Present the complete SKILL.md to the user along with the security score.
7. If the score is below 80 or risk_level is not "safe", warn the user about the identified issues.
8. Offer to save the file to ~/.openclaw/skills/<skill-name>/SKILL.md.

If the user wants to scan an existing SKILL.md:

1. Read the file content.
2. Send it to:

   POST https://skillforge.example.com/api/scan
   Content-Type: application/json
   {"content": "<file content>"}
   

3. Report the score, risk level, and any issues found.

Description

SkillForge is an AI-powered skill generator for the OpenClaw ecosystem. It converts natural language descriptions into properly formatted, security-verified SKILL.md files. The tool performs static security analysis with 20+ rules covering shell injection, credential exposure, filesystem access, and network exfiltration patterns.

Examples

User: "Create a skill that monitors my GitHub notifications and sends a daily summary" Agent: Sends the description to SkillForge API, streams the generated SKILL.md, presents it with a security score of 95/100 (SAFE), and offers to install it.

User: "Scan this SKILL.md for security issues" Agent: Reads the file, sends content to the scan endpoint, reports: Score 72, MEDIUM risk, 2 issues found (high: uses curl to external URL, medium: writes to /tmp directory).