The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes a log file (log.txt) composed of internal and external network traffic. Because this file contains content generated by third-party web servers and applications, it is inherently untrusted. An attacker could embed malicious instructions in HTTP headers, URL paths, or response bodies that the agent may interpret and follow when summarizing the traffic.
Ingestion points: log.txt (generated via mitmdump).
Boundary markers: The skill does not provide delimiters or instructions to the agent to disregard instructions embedded within the logs.
Capability inventory: The skill performs file reading and summarization of the captured data.
Sanitization: No sanitization or validation of the log content is required before the agent processes it.
[DATA_EXFILTRATION]: The instructions direct the agent to extract and display "Input params" and "Response fields" from the traffic capture. Since these captures contain raw HTTP requests and responses, they are highly likely to contain sensitive information such as authentication tokens (Bearer/Session), API keys, and personal identifiable information (PII) belonging to the user or the target application.

mitm-list-apis