The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides templates that ingest untrusted data from network request components (URL parameters, headers, and cookies) and uses them to perform sensitive actions like network requests.
Ingestion points: The handleProxy function in templates/edge-function.ts extracts a targetUrl from query parameters, and templates/edge-middleware.ts parses cookies and headers from incoming requests to make routing decisions.
Boundary markers: The templates do not use delimiters or include instructions to the agent to ignore potentially malicious embedded directives in the processed request data.
Capability inventory: The skill utilizes the fetch API for outbound network communication and interacts with KVNamespace for reading and writing persistent storage data.
Sanitization: The handleProxy implementation in templates/edge-function.ts performs minimal validation, checking only the URL protocol (http: or https:) but failing to implement a whitelist or block internal network addresses (e.g., local loopback or cloud metadata services), facilitating SSRF attacks.
[DATA_EXFILTRATION]: The handleProxy implementation in templates/edge-function.ts contains a pattern that forwards sensitive credentials to an arbitrary, user-controlled destination.
Evidence: The template explicitly forwards the Authorization header from the incoming request directly to the external targetUrl via the fetch call, which can lead to the exfiltration of authentication tokens to attacker-controlled infrastructure.

edge-computing-patterns