We take the security of Trust Gateway seriously. If you believe you have discovered a vulnerability, security flaw, or cryptographic weakness, please do not open a public GitHub issue.

Instead, report vulnerabilities privately through one of the following channels:

• Security Advisory Form: Submit a report via GitHub Private Vulnerability Reporting on this repository.
• Email: Reach out directly to security@lianxi.io with details of the issue.

To help us investigate and resolve issues quickly, please include:

1. Description: Clear description of the vulnerability, including affected components or protocol flows.
2. Reproduction Steps: Step-by-step instructions, script, or proof-of-concept payload demonstrating the issue.
3. Impact: Potential impact of the flaw (e.g., grant forgery, bypass of argument binding, key exposure).
4. Environment: Version or commit hash tested, operating system, and deployment topology.

Trust Gateway enforces cryptographic and physical execution control boundaries for AI agents. The following invariants are P0 security properties:

1. No Direct Execution: Agents never hold standing credentials and cannot execute mutations directly.
2. Cryptographic Argument Binding: Every ExecutionGrant is bound to a SHA-256 input_hash of exact arguments.
3. Single-Use Nonces: Grants are single-use (jti nonce tracking) and short-lived.
4. Asymmetric Signing: In production (LIANXI_ENV=production), grants must be signed with Ed25519 keypairs. HMAC symmetric keys are strictly gated to development environments.
5. Fail-Closed Default: Undefined tools, invalid signatures, missing policies, or unverified claims result in immediate execution denial.

• Initial Response: Within 48 hours acknowledging receipt of the report.
• Triage & Status: Updates provided within 5 business days detailing vulnerability severity and remediation timeline.
• Fix & Disclosure: Coordinated disclosure after patches are published.