A vetted skills registry for AI coding agents (claws) curated by FeedbackPulse EX Platform
Open a PR with your proposed skill.
Every submission is reviewed by multiple frontier AI models before human approval:
| Agent | Focus |
|---|---|
| Claude Opus 4.6 | Prompt injection, instruction manipulation |
| GPT-5.2 | Code execution, system access risks |
Skills are checked for:
- Prompt injection (hidden instructions, "ignore previous", encoded commands)
- Dangerous tool usage (unrestricted Bash, rm -rf, sensitive file access)
- Data exfiltration (network calls to external URLs, reading credentials)
- Social engineering or manipulation attempts
- Overly permissive allowed-tools
A skill must pass both AI reviews, then receive human approval to be merged.
Note: This repository contains Anthropic's implementation of skills for Claude. For information about the Agent Skills standard, see agentskills.io.