The Agent Skills Directory

[COMMAND_EXECUTION]: The helper script scripts/agent.py facilitates command execution by wrapping the velociraptor binary using subprocess.run. It allows users to execute VQL (Velociraptor Query Language) queries and collect artifacts. The implementation uses list-based arguments for subprocess.run to prevent shell injection, though the underlying binary possesses significant system capabilities.
[EXTERNAL_DOWNLOADS]: The documentation in SKILL.md provides instructions for downloading official binaries from the Velocidex GitHub repository. This is a standard deployment procedure for the tool and targets a well-known security project.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface, as it is designed to ingest and process data from endpoint forensic collections which could contain attacker-controlled content.
Ingestion points: Forensic data and query results returned by the velociraptor binary and handled by scripts/agent.py (via run_query and collect_artifact).
Boundary markers: The script does not implement specific delimiters or 'ignore' instructions for the data retrieved from endpoints.
Capability inventory: The skill has the capability to execute the velociraptor binary, write files to the local system (zip results), and perform network communication via the Velociraptor protocol.
Sanitization: The script outputs raw data from the binary to stdout/stderr without sanitization or filtering intended for LLM context safety.

fleet-hunting-with-velociraptor