⚠️ This file is project documentation, not part of the source code.

RACT executes user-provided intents, writes files, and can run shell commands. We take security seriously.

If you discover a security vulnerability — including prompt injection, arbitrary file write/escape, sandbox escape, credential leakage, MCP tool abuse, or any issue that could compromise a user's machine or data — please report it privately.

• Email: security@lucasroot.com
• Subject: [RACT Security] <short description>
• PGP: security@lucasroot.com key fingerprint is published at https://lucasroot.com/pgp/security.asc. Please encrypt sensitive details.

Please include:

• A clear description of the vulnerability.
• Steps to reproduce, if applicable.
• The version of RACT you are using.
• Any suggested mitigation or fix.

We will acknowledge receipt within 72 hours and share a timeline for a fix. We will publicly disclose the issue with appropriate credit once a patch is available, unless you request otherwise.

• RACT classifies every action into a capability tier (T0–T3) and refuses T3 actions by default.
• RACT writes artifacts only within the configured project directory; absolute paths and paths resolving outside the workspace are refused.
• RACT refuses reads of sensitive files (.env, *.pem, *.key, ~/.ssh/**, etc.).
• RACT refuses package-registry publishes, rm -rf on untracked paths, and full-workspace uploads above the chunk threshold.
• The --yolo and --allow-tier-3 flags disable safety checks. Use them only in trusted environments.
• Provider API keys should be provided via environment variables or secure config, never committed to version control.
• The SafetyGuardrail blocks known dangerous patterns, but it is not a substitute for reviewing generated code before execution.

RACT is licensed under the PolyForm Noncommercial License 1.0.0.

Dr. Lucas Root, Ph.D.