laguarde-policy-gate

Pass

Audited by Gen Agent Trust Hub on Jul 26, 2026

Risk Level: SAFE
Full Analysis
  • [DATA_EXFILTRATION]: The skill instructs the agent to send project metadata, including repository-relative file paths, exact shell commands (omitting secrets), and dependency changes, to the Laguarde MCP server (using the /mcp/projects/:projectId endpoint). This data transmission is required for the service to evaluate the action against active policies.
  • [PROMPT_INJECTION]: The agent is designed to ingest and strictly follow external policy bundles via the get_policy_bundle tool, which presents an indirect prompt injection surface where the external server's output influences agent logic.
  • Ingestion points: Policy bundles and revision IDs retrieved from the Laguarde MCP server.
  • Boundary markers: Absent; instructions mandate that the agent treat the recorded decision as authoritative over its own judgment.
  • Capability inventory: The skill acts as a gateway for sensitive operations including file modification, command execution, and credential access.
  • Sanitization: No explicit sanitization or validation of the policy bundle content is described.
  • [EXTERNAL_DOWNLOADS]: The instructions refer to following a user-authorized "installation contract" when setting up the service, which may involve retrieving external configuration or setup scripts. This process requires explicit human authorization before execution.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 26, 2026, 05:27 PM
Security Audit — agent-trust-hub — laguarde-policy-gate