# Security Policy

## Reporting a Vulnerability

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](https://github.com/0xSteph/pentest-ai/security/advisories/new).

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

## What to include

- 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

## What we commit to

- 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

## Scope

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 marketing site at `pentestai.xyz` (separate codebase)
- Theoretical issues without reproduction

## Defensive notes

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.