Cc Skill Audit — Security skill for Claude Code

cc-skill-audit

**Status (2026-09-11):** Frozen at 0.2.0. Daily path for auditing a local skill folder is the Claude Code skill `/audit-skill`. This CLI / PreToolUse scanner is the last published standalone release; no new detectors are planned here.

Security scanner & PreToolUse firewall for Claude Code third-party skills.

    

[繁體中文 README](README.zh-TW.md)

Detects undisclosed telemetry, outbound data exfiltration, obfuscated code, binary blobs, and suspicious patterns **before** you install a third-party skill.

Born from a [real incident](https://blog.chibakuma.com/ai-audit-gstack-telemetry-2/) where a popular Claude Code skill pack was found to silently log project names — even with telemetry set to "off."

Why

Claude Code skills run with **full OS-level permissions**. There is no sandbox, no permission model, no review process. Installing a skill = giving it access to your repos, SSH keys, env files, and everything else.

In Q1 2026 alone:

• A popular skill pack logged repo names without consent (our incident)
• Trivy VS Code extension (964K installs) was injected with malicious AI prompts
• Cline CLI was compromised via stolen npm token
• Multiple AI browser extensions were caught harvesting chat histories

cc-skill-audit is a simple, local-only, zero-dependency tool that scans skills for red flags before you install them.

Quick Start

git clone https://github.com/MakiDevelop/cc-skill-audit.git
cd cc-skill-audit
./install.sh

Or scan without installing:

./bin/cc-skill-audit /path/to/suspicious-skill

What it scans

| Category | What i