Do not open a public issue for security vulnerabilities.

Use GitHub private vulnerability reporting: navigate to the Security tab on leocelis/eif and click Report a vulnerability. This creates a private advisory visible only to the maintainer, and keeps the report private until a fix is coordinated.

Please include:

1. Description: what the vulnerability is and what it allows
2. Steps to reproduce: minimal reproducible case
3. Affected versions
4. Impact: confidentiality, integrity, or availability

This policy covers:

• The eif-engine Python library (eif/)
• The MCP servers (eif/mcp_server/): stdio (eif-mcp-server) and local HTTP (eif-mcp-http-server)
• The P1 code-execution sandbox (eif/falsify/native_tools.py): AST validation, subprocess isolation

Out of scope:

• Third-party dependencies (report to their maintainers; transitive CVEs are tracked via Dependabot)
• Forks or derivative works not maintained in this repository
• The security of your own deployment infrastructure (see below)

EIF is self-hosted software. You own the security of your deployment.

• The HTTP server is intended for local use. Run it on localhost or inside a private network. If you expose it beyond that, set EIF_API_KEY, terminate TLS in front of it, and restrict access at the network layer.
• Keys stay out of source control. Never commit .env, EIF_API_KEY, or OPENAI_API_KEY. Copy .env.example to .env and keep .env gitignored.
• The core engine makes no outbound network calls. Optional LLM-backed helpers only activate when OPENAI_API_KEY is set; unset it for fully offline operation.
• The P1 sandbox executes generated Python in a subprocess after AST-based validation. It is designed for EIF's own stdlib-only code templates. Do not route untrusted arbitrary code through it.