The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted log data through scripts/process.py and scripts/agent.py, creating a potential surface for indirect prompt injection. 1. Ingestion points: External JSON and CSV log files are ingested by scripts/process.py (parse_logs) and scripts/agent.py (parse_splunk_results). 2. Boundary markers: No explicit delimiters or instructions to ignore embedded content are included in the generated hunt_report.md or JSON output. 3. Capability inventory: The skill performs local file system writes and generates console output intended for the AI agent's review. 4. Sanitization: While the scripts use structured parsing (JSON/CSV) and regex matching for detection, the resulting reports interpolate log-provided data like account names and hostnames directly into the output without sanitization for downstream LLM safety.

detecting-lateral-movement-with-splunk