agentsclimarketplace
Sign in

SKA auditor

Skill asong56/skills/08-extend/SKA-auditor

Security audit for Claude Code SKILL.md files using static analysis + LLM-powered threat detection. Detects prompt injection, data exfiltration, MCP tool poisoning, privilege escalation, sandbox escapes, and 15 other threat categories. Run before ...From its SKILL.md

Install
npx -y skills add asong56/skills --skill SKA-auditor

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.0 KB, 616 tokens by cl100k_base, as published. Nobody here has run it

/SKA — Skill Security Auditor

Use this skill to run a security audit on .claude/skills/ directories or individual SKILL.md files before installation. Combines static YARA-based pattern matching with LLM semantic analysis via a LangGraph multi-node pipeline.

Setup

cd .claude/skills/SKA/scripts
pip install -e . --break-system-packages

Requires: Python 3.11+, ANTHROPIC_API_KEY (or OPENAI_API_KEY) in environment.

Usage

# Scan a single skill file
python -m SKA path/to/SKILL.md

# Scan a full skills directory
python -m SKA .claude/skills/

# JSON output (for CI integration)
python -m SKA .claude/skills/ --output json

# Exit 1 if any HIGH or CRITICAL finding (for pre-install gate)
python -m SKA .claude/skills/ --fail-on HIGH

What It Detects

SKA runs 14 static analyzers + 3 LLM semantic analyzers in parallel:

CategoryExamples
Prompt injectionOverride system prompt, jailbreak instructions
Data exfiltrationLeak env vars, files, secrets via tool calls
MCP tool poisoningManipulate tool definitions to cause unsafe ops
Privilege escalationGain capabilities beyond skill scope
Rogue agent patternsOverride agent behavior, autonomous takeover
Supply chain attacksMalicious dependencies in skill setup steps
Memory poisoningCorrupt cross-session memory/context
Excessive agencySkills claiming unlimited permissions
Harmful contentInstructions for harmful real-world actions
System prompt leakageExtract confidential system instructions
Output handling abuseManipulate Claude output to downstream systems
Cryptominer patternsEmbed crypto-mining in skill execution
MCP rug-pullTools that change behavior after trust established
Least-privilege violationsOver-broad tool permission requests

Risk Levels

  • LOW — Safe to install
  • MEDIUM — Review flagged sections before installing
  • HIGH — Do not install without manual security review
  • CRITICAL — Likely malicious — reject

CI Integration

Add to your pre-skill-install workflow:

- name: Scan skill for security
  run: |
    cd .claude/skills/SKA/scripts
    pip install -e . -q
    python -m SKA ../new-skill/ --fail-on MEDIUM

SARIF Output

For GitHub Advanced Security integration:

python -m SKA .claude/skills/ --output sarif > results.sarif

Source

Powered by the SKA Python package in scripts/. Architecture: LangGraph pipeline → static analysis nodes → LLM semantic nodes → meta-aggregator → report.

What ships with it: 59 files

394.0 KB alongside SKILL.md, 51 of them executable

scripts/

19 more files not listed here. See all 59 in the repository.

Keep looking

Skills are one crate of 326,144. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.