The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides an implementation example for a 'BuddyAgent' that processes untrusted data from incoming HTTP requests and interpolates it directly into an AI model prompt. This creates a surface for indirect prompt injection where an attacker could provide malicious instructions via the 'message' field in the request payload.
Ingestion points: The body.message variable is extracted from request.json() in the BuddyAgent implementation within SKILL.md.
Boundary markers: Absent. The untrusted input is passed directly to the model as a user message without additional delimiters or instructions to ignore embedded commands.
Capability inventory: The script has the capability to perform AI inference via env.AI.run and query a vector database via env.VECTORIZE.query.
Sanitization: There is no evidence of sanitization, filtering, or validation of the input message before it is processed by the AI model.

cloudflare-workers