cerbos-pep-integration
Pass
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides detailed guidance for securing AI agent tool calls and MCP servers through policy enforcement.
- Ingestion points: Processes data from MCP tool call arguments and user-provided agent instructions as described in
SKILL.md. - Boundary markers: Recommends using the human user's identity (principal) rather than the agent's identity to ensure agents cannot perform actions unauthorized for the user.
- Capability inventory: Enforcement is performed at the tool-call boundary using
checkResourcecalls to the Cerbos PDP before any tool execution occurs. - Sanitization: Encourages the use of
list_toolsfiltering to hide unauthorized capabilities from the model's view while maintaining server-side enforcement as the primary security gate. - [EXTERNAL_DOWNLOADS]: References official Cerbos SDKs and documentation from
github.com/cerbos,docs.cerbos.dev, and official package registries (NPM, PyPI, Maven, NuGet, Crates.io, Composer, Gem). - [COMMAND_EXECUTION]: Provides standard Docker and CLI examples for running a local Cerbos PDP instance for development purposes.
Audit Metadata