cerbos-synapse-extension

Pass

Audited by Gen Agent Trust Hub on Aug 25, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The Python WASM build pipeline in references/shared/python-wasm-common.md fetches wasm-tools from the Bytecode Alliance, binaryen from the WebAssembly organization, and extism-py from the Extism project. These are established organizations and official repositories for WebAssembly development tooling. \n- [PROMPT_INJECTION]: The skill facilitates the development of extensions that process untrusted external data (such as HTTP headers, request bodies, and external data source results) to determine authorization effects. This functionality represents an indirect prompt injection surface. \n
  • Ingestion points: Extensions process data from request.header, request.body.json, and cerbos.data_source_lookup as seen in references/starlark-proxy-extension.md and references/starlark-route-extension.md. \n
  • Boundary markers: The framework uses structured data types (Starlark structs and JSON) which provide inherent boundaries for data processing. \n
  • Capability inventory: Extensions are granted capabilities to perform network requests, manage local caches, and invoke the Cerbos Policy Decision Point (PDP). \n
  • Sanitization: The documentation guides users to implement explicit mapping logic and use standard JSON parsing for data handling. \n- [COMMAND_EXECUTION]: The references/shared/run-and-test.md file provides standard docker and docker compose command examples to help developers run and test their extensions within the Synapse server environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 25, 2026, 01:11 AM
Security Audit — agent-trust-hub — cerbos-synapse-extension