The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted API gateway access logs that could contain malicious instructions designed to deceive the AI agent during the log review process.
Ingestion points: The scripts/agent.py file reads external data from JSON or CSV log files via load_api_logs (using pandas.read_json and pandas.read_csv).
Boundary markers: While the output is formatted as a structured JSON report, there are no boundary markers or specific instructions provided to the agent to treat the analyzed data as untrusted or to ignore embedded instructions.
Capability inventory: The script performs file system read/write operations and utilizes the pandas library for data processing.
Sanitization: The script extracts and re-shares potentially malicious request bodies and query strings into its final report without any sanitization, filtering, or escaping, which could lead to the agent executing instructions contained within the logs.

analyzing-api-gateway-access-logs