The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill implements a tenant-aware reverse proxy that processes untrusted data from incoming HTTP requests and forwards it to an external API, creating a potential surface for indirect injection if the processed data is later interpreted by an agent.
Ingestion points: Untrusted data enters via req.params.tenant, the wildcard path segment req.params[0], req.originalUrl query strings, and the raw request body in references/proxy.ts.
Boundary markers: The proxy implements an allow-list check using permittedPrefixes to restrict reachable upstream paths, but it does not apply delimiters or specific "ignore" instructions to the data payloads being forwarded.
Capability inventory: The skill possesses network communication capabilities, using the fetch API to forward requests to the iGrant.io OWS environment.
Sanitization: The implementation performs path validation against the allow-list and method validation (GET, POST, PUT, DELETE), but it forwards request bodies verbatim as a Uint8Array without content inspection or sanitization.

igrantio-backend-proxy