The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads Velociraptor binaries from the official GitHub repository for the project (Velocidex/velociraptor). This is a standard and expected deployment method for this security tool.
[COMMAND_EXECUTION]: Deployment instructions include the use of administrative commands such as sudo for service installation, wget for fetching binaries, and docker run for containerized deployment. These operations are consistent with the skill's primary purpose of system administration and incident response setup.
[PROMPT_INJECTION]: The skill exhibits a vulnerability surface for indirect prompt injection (Category 8) because its core functionality involves processing untrusted forensic data (such as event log entries, process names, and command lines) retrieved from endpoints.
Ingestion points: The scripts/process.py script reads and analyzes JSON-formatted forensic result files in the analyze_collection_results function.
Boundary markers: Absent; the script processes external data without using delimiters or specialized markers to distinguish it from the agent's internal logic.
Capability inventory: The toolset includes file system write operations (json.dump in process.py) and network request capabilities (requests.get in scripts/agent.py).
Sanitization: While the script performs basic substring matching for known malicious indicators (e.g., 'mimikatz'), it does not implement sanitization or escaping of the ingested data to prevent it from potentially influencing the agent's behavior if the output is processed by an LLM.

implementing-velociraptor-for-ir-collection