TrustFabric is pre-audit software. It has not yet undergone an independent security review. Do not rely on it as your sole control in a production security boundary until an audit is completed. (Tracking: a third-party audit is a prerequisite for the 1.0 release.)

Please do not open a public issue for security reports. Instead, email the maintainer (see repository profile) with:

• a description of the issue and its impact,
• reproduction steps or a proof of concept,
• any suggested remediation.

You will receive an acknowledgement within 72 hours. We ask for a 90-day coordinated disclosure window.

In scope: the enforcement gates (identity, permission, delegation, audit), both transports (STDIO interceptor, HTTP reverse proxy), policy parsing, and the audit chain's tamper-evidence properties.

Out of scope: vulnerabilities in wrapped MCP servers themselves, and denial-of-service against the example servers (they are test fixtures).

• No token revocation: a valid token works until it expires. Use short TTLs.
• SSE-streamed list responses are passed through unfiltered (JSON list responses are filtered).
• Policy constraints are glob/equality matches; they cannot express relational rules.
• The HS256 mode is for development; anyone holding the shared secret can mint tokens. Use RS256/ES256 in any real deployment.
• prompts/list and prompts/get are currently passed through ungated. Prompts are static templates (lower exfiltration risk than tools or resources), but gating them is planned; do not store secrets in prompts.