We take security vulnerabilities seriously. If you discover a security issue in this project, please report it responsibly.

Report vulnerabilities through GitHub Private Vulnerability Reporting. This creates a private advisory visible only to repository maintainers until a fix is available.

For sensitive disclosures or if GitHub PVR is unavailable, contact us via our website.

We follow a 90-day coordinated disclosure timeline:

1. Day 0: Vulnerability reported.
2. Day 1-2: Report acknowledged.
3. Day 1-5: Initial assessment and severity classification.
4. Day 5-90: Fix developed, tested, and released.
5. Day 90: Public disclosure, regardless of fix status (with reporter agreement).

If we need additional time beyond 90 days, we will communicate this to the reporter with a clear justification and revised timeline.

The following components are covered by this security policy:

• Skill definitions (skills/security-audit/SKILL.md, reference documents)
• Scripts (scripts/check_risky_command.py, scripts/security-audit.sh)
• Hooks (hooks/hooks.json, PreToolUse hook configuration)
• CI/CD workflows (.github/workflows/)
• Package metadata (composer.json)

The following are NOT covered by this security policy:

• Third-party dependencies (report to upstream maintainers)
• The AI agent itself (Claude, Cursor, Copilot -- report to the respective vendor)
• User-generated content from using the skill in audits
• Infrastructure where the skill is deployed (your local machine, CI runners)

This project implements the following security measures:

A PreToolUse hook (scripts/check_risky_command.py) inspects commands before execution, detecting patterns associated with:

• Destructive file operations (rm -rf /, chmod 777)
• Network exfiltration (curl | sh, wget with suspicious targets)
• Credential exposure (hardcoded secrets in commands)
• Privilege escalation (sudo with dangerous arguments)

Reference documents and audit scripts are aligned with:

• OWASP Top 10 (2021)
• CWE Top 25
• OWASP Application Security Verification Standard (ASVS)

• All PHP code examples use declare(strict_types=1).
• Vulnerable code examples are clearly marked with // VULNERABLE - DO NOT USE.
• Secure alternatives are provided with // SECURE: annotations.

• Hook integrity: The PreToolUse hook script (scripts/check_risky_command.py) has no post-installation integrity verification. If an attacker gains write access to the installed skill directory, they could modify the hook to suppress warnings. Users should monitor file modifications in their skill installation directory.
• Hook scope: The PreToolUse hook only intercepts the Bash tool. Commands executed via other tool types (e.g., writing a script file and executing it separately) are not inspected by the hook.

We credit responsible reporters in release notes (unless anonymity is requested). If you report a valid vulnerability, we will:

• Acknowledge your contribution in the release notes for the fix.
• Add your name (or alias) to a CONTRIBUTORS section if desired.
• Provide a reference letter or public acknowledgment upon request.

For non-vulnerability security questions about this project, open a GitHub Discussion.