Skip to content

Security: ledjay/skills

Security

SECURITY.md

Security Policy

Reporting a Vulnerability

If you discover a security vulnerability in this repository, please report it privately:

Do not open a public issue for security vulnerabilities.

Security Measures

This repository implements multiple layers of security to protect against malicious skills:

1. Pre-commit Hooks (Lefthook)

# Install hooks
lefthook install

# Run manually
lefthook run pre-commit

Checks:

  • security-scan.py - Detects prompt injection, malicious code, data exfiltration
  • gitleaks - Detects hardcoded secrets
  • unicode-check.py - Detects Unicode smuggling attacks

2. CI/CD Scanning (GitHub Actions)

Every pull request is automatically scanned:

  • Cisco Skill Scanner - Multi-engine detection (static + LLM + dataflow)
  • Local security scan - Pattern-based detection
  • Unicode check - Invisible character detection
  • Gitleaks - Secret detection

Results appear in the Security tab.

3. Threat Detection

We detect the following threat categories:

Category Examples Severity
Prompt Injection "ignore previous instructions", DAN, jailbreak 🔴 Critical
Data Exfiltration curl to unknown domains, file uploads 🔴 Critical
Malicious Code eval(), exec(), reverse shells 🔴 Critical
Unicode Smuggling Invisible Unicode Tags, zero-width chars 🔴 Critical
Obfuscation Base64, hex encoding, fromCharCode 🟡 High
Credentials API keys, tokens, passwords 🟡 High
Suspicious URLs Unknown domains, webhook endpoints 🟢 Medium

4. Known Attack Patterns

Based on Snyk's ToxicSkills research and Cisco's threat intelligence:

ClawHavoc Campaign Patterns:

  • Metadata poisoning with overbroad descriptions
  • Credential theft via embedded scripts
  • Atomic macOS Stealer (AMOS) payloads
  • VMProtect-packed infostealers

Prompt Injection Variants:

  • Instruction hierarchy overrides
  • Unicode Tag smuggling (U+E0000-U+E007F)
  • Zero-width character injection
  • Base64/hex encoded instructions

Data Exfiltration Methods:

  • Curl to attacker-controlled servers
  • Multipart form uploads
  • Webhook exfiltration
  • Environment variable snooping

Installing Security Tools

Required (for contributors)

# Install Lefthook and Gitleaks
brew install lefthook gitleaks

# Initialize hooks
lefthook install

Optional (enhanced scanning)

# Cisco Skill Scanner (Python 3.10+)
pip install cisco-ai-skill-scanner

# Run full scan
skill-scanner scan ./ --format text

Security Best Practices for Skills

When contributing skills, follow these guidelines:

✅ DO

  • Keep SKILL.md under 500 lines
  • Use clear, specific descriptions
  • Document all scripts and their purpose
  • Use relative paths, not absolute
  • Validate all user inputs

❌ DON'T

  • Don't include hardcoded credentials
  • Don't use eval(), exec(), or dynamic code execution
  • Don't make network requests to unknown domains
  • Don't hide instructions in comments or encoded strings
  • Don't use Unicode tricks or zero-width characters

False Positives

If a security scan produces false positives:

  1. Check if the pattern is legitimate (e.g., example code)
  2. Add to .gitleaks.toml allowlist if it's a secret false positive
  3. Document in the skill's README if it's a known pattern

References

There aren’t any published security advisories