The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input which could contain malicious instructions designed to influence downstream modules or file system operations.
Ingestion points: The user_request field in the '接收 (来自 Hub)' section, which is passed in full without summarization.
Boundary markers: The file does not specify delimiters or boundary markers to isolate the untrusted user_request content from the system instructions.
Capability inventory: The skill coordinates several downstream workers (afa-creative, afa-fb, afa-google, afa-tiktok) and writes files to the ./brand-brain/ directory.
Sanitization: No sanitization, escaping, or filtering mechanisms are described for the incoming user_request data.

afa-paid