We release patches for security vulnerabilities in the following versions:

We take the security of AI Agent Skills seriously. If you discover a security vulnerability, please follow these steps:

Please do not report security vulnerabilities through public GitHub issues.

Open a private security advisory on GitHub:

Report here: Create a private security advisory

Do not email secrets or exploit details to public channels.

Please include the following information:

• Type of vulnerability (e.g., code injection, path traversal, unsafe installer behavior)
• Full paths of source file(s) related to the vulnerability
• Location of the affected source code (tag/branch/commit or direct URL)
• Step-by-step instructions to reproduce the issue
• Proof-of-concept (if possible; keep payloads minimal)
• Impact of the vulnerability (what an attacker could do)

• Initial Response: Within 48 hours
• Status Update: Within 7 days
• Fix Timeline: Depends on severity
  • Critical: Within 7 days
  • High: Within 14 days
  • Medium: Within 30 days
  • Low: Next release cycle

When using AI Agent Skills:

1. Review Skills Before Use

   • Read skill content before installation
   • Check for suspicious commands or patterns
   • Verify source attribution (origin:, source_url:, NOTICE)

2. Prefer curated presets over --all

   • Start with --preset minimal or --preset developer
   • Dual-use / red-team skills are opt-in (see below)

3. Keep Updated

   • Use ./scripts/update.sh regularly
   • Check for security advisories
   • Follow release notes

4. Validate Installation

   • Run ./scripts/install.sh --validate-only or ./scripts/validate.sh
   • Check for unexpected files after install

5. Sandbox Testing

   • Test new skills in isolated environments first
   • Use containers or VMs for untrusted skills
   • Monitor system behavior

When contributing skills:

1. No Secrets

   • Never include API keys, tokens, or passwords
   • Use environment variables for sensitive data
   • Check commits for accidental leaks

2. Input Validation

   • Validate all user inputs
   • Sanitize file paths
   • Escape shell commands

3. Safe Defaults

   • Fail securely by default
   • Require explicit confirmation for destructive actions
   • Use least privilege principle

4. Dependencies

   • Minimize external dependencies
   • Pin dependency versions where practical
   • Audit dependencies regularly

Some skills document security research, red-teaming, or dual-use workflows (for example skills/red-teaming/godmode and certain research/devops skills). They are instruction files, not exploits shipped as runnable malware, but they can still steer an agent toward unsafe actions if used carelessly.

Guidance:

• Review the skill body before install
• Prefer agent-level allowlists / confirmation gates
• Install dual-use content only when you explicitly need it (e.g. --category red-teaming or --preset agentic)
• Do not use these skills to attack systems you do not own or lack authorization to test

Skills are markdown files with instructions for AI agents. They do not execute code by themselves, but:

• AI agents may execute commands based on skill instructions
• Users should review what skills instruct agents to do
• Malicious or dual-use skills could instruct agents to perform harmful actions

scripts/install.sh, scripts/install.py, scripts/update.sh, and scripts/uninstall.sh write to local directories you choose (defaults under ~/.hermes/skills, ~/.claude/skills, or ~/.cursor/skills). Always inspect flags (--dry-run, --validate-only, --target) before running.

1. Review Before Use: Always read skill content
2. Attribution: Prefer skills with complete source_url / source_license metadata
3. Validation: Use the validation scripts to check skill integrity
4. Agent Safeguards: Keep agent tool permissions constrained where possible

Security updates are released as:

• Patch versions (e.g., 1.7.2) for security fixes
• GitHub Security Advisories for critical issues
• CHANGELOG.md entries marked with [SECURITY]

Subscribe to:

• GitHub Security Advisories
• Release notifications

• Skill content that could lead to unsafe agent-driven actions
• Installation scripts (install.sh, install.py, update.sh, uninstall.sh, validate.sh)
• Documentation that could mislead users into unsafe practices
• Dependencies declared by this repo's tooling

• AI agent behavior (report to agent maintainers)
• Third-party tools referenced in skills
• User misconfiguration
• Social engineering attacks
• Unauthorized use of dual-use skills against third parties

We believe in responsible disclosure and will credit security researchers who report vulnerabilities (unless they prefer to remain anonymous).

For security questions that are not vulnerabilities, please open a regular GitHub issue.

Thank you for helping keep AI Agent Skills secure.