The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external SIEM API responses which could contain malicious content designed to influence the agent's behavior during report analysis.
Ingestion points: The check_siem_detection function in scripts/agent.py ingests JSON alert data from a remote SIEM service.
Boundary markers: There are no specific delimiters or protective instructions used to isolate the ingested alert data from the agent's instruction context.
Capability inventory: The skill possesses network communication capabilities (requests) and file-writing capabilities (json.dump in scripts/agent.py and pandas.to_csv in scripts/process.py).
Sanitization: While the script parses JSON structure, it does not sanitize the textual content of the alerts before incorporating them into reports that the AI agent may later process.
[DATA_EXFILTRATION]: scripts/agent.py performs network GET requests to user-supplied targets and configured SIEM endpoints, which represents network operations to non-whitelisted external domains.

implementing-continuous-security-validation-with-bas