Horton is a local training tool. It:

• Runs entirely on your machine
• Makes outbound API calls only to your chosen LLM provider (Anthropic or Google)
• Stores nothing remotely — no telemetry, no session data, no conversation logging
• Never handles real credentials — tokens generated during gameplay are random, session-scoped, and discarded at exit
• Has no web interface, no network listener, no authentication surface

The attack surface is intentionally minimal by design.

If you find a security issue in Horton itself (e.g., a path traversal in scenario loading, a prompt that causes the tool to write files outside its directory, a dependency with a known CVE):

1. Do not open a public GitHub issue
2. Email a description to the maintainer (visible on the GitHub profile)
3. Include: what you found, how to reproduce it, and your assessment of impact
4. You'll get a response within 5 business days

We take supply chain hygiene seriously — the scenario lore exists because we do.

All runtime dependencies are pinned in requirements.txt. To audit:

pip install pip-audit
pip-audit -r requirements.txt

This tool is designed for authorized security training environments only. See README.md for the full responsible use policy. Using social engineering techniques against real people without their explicit consent and authorization is illegal in most jurisdictions and unethical in all of them.