If you find a security issue in the pentest-ai CLI / MCP server, please report it privately. Do not open a public issue, do not post on social media, do not include the issue in a pull request title.

Preferred channel: open a GitHub Security Advisory at github.com/0xSteph/pentest-ai/security/advisories/new.

Backup: email security@pentestai.xyz with the subject [SECURITY] short description.

• Affected file (path) and entry point (ptai, pentest-ai, MCP server, etc.)
• Affected version (pyproject.toml / ptai --version, and CHANGELOG.md)
• Reproduction steps, including command line and environment
• Impact you observed or believe likely
• Any suggested fix

• Acknowledge receipt within 3 business days
• Initial triage within 7 business days
• Coordinated disclosure window of up to 90 days for confirmed vulnerabilities
• Credit in release notes if you want it

In scope:

• Python source (agents/, engine/, cli/, mcp_server/)
• Tool wrappers (tools/)
• Playbooks (playbooks/)
• Authentication code (auth_handler, auth_session, JWT logic)
• Scope validation and target expansion
• Findings DB and evidence handling

Out of scope:

• Third-party tools invoked by the engine (nmap, nuclei, sqlmap, etc.) — report upstream
• The SaaS platform at app.pentestai.xyz (separate codebase, separate disclosure)
• The marketing site at pentestai.xyz (separate codebase)
• Theoretical issues without reproduction

This is offensive security tooling. It executes real network and host operations. Engagements MUST run with authorized scope. If you find a way to escape scope enforcement, exfiltrate credentials beyond the engagement boundary, or coerce the engine into running outside the declared target list, treat that as a high-severity finding.