The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external, untrusted Windows Security Event Logs (EVTX files) which can include attacker-controlled data like target usernames.
Ingestion points: scripts/agent.py loads and parses the EVTX log file provided via command-line arguments.
Boundary markers: No boundary delimiters or system prompt instructions are present to prevent an upstream language model from executing commands embedded in log fields.
Capability inventory: The script executes local file system writes to save the JSON detection report and outputs results to stdout.
Sanitization: Extracted fields such as TargetUserName are parsed via standard XML parsing and appended directly into the report dictionary without input filtering or escaping.

detecting-rdp-brute-force-attacks