threat-hunting

Pass

Audited by Gen Agent Trust Hub on Sep 19, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is explicitly designed to process external and untrusted log data from various sensors and SIEM platforms, which presents a surface for indirect prompt injection.
  • Ingestion points: Data enters the agent's context through log files such as events.ndjson and conn.log, as well as via SIEM exports from Splunk, Sentinel, and Elasticsearch as described throughout the skill and its payload guides.
  • Boundary markers: The provided example scripts and templates for log ingestion (e.g., in hunt_pipeline.py) do not implement specific boundary delimiters or "ignore embedded instructions" warnings for the ingested fields.
  • Capability inventory: The skill allows access to Bash, Read, Write, and Edit tools. If the agent processes a log entry containing adversarial instructions, these tools could be leveraged to perform unauthorized actions.
  • Sanitization: The skill focus is on pattern matching and statistical analysis (e.g., beacon detection) but does not demonstrate explicit sanitization or filtering of the raw log content to prevent injection attacks.
  • [EXTERNAL_DOWNLOADS]: The skill facilitates the download and installation of various industry-standard security tools and dependencies.
  • Fetches the sigma-cli utility and its associated plugins from the official package registry.
  • Recommends cloning the official SigmaHQ rule repository from its well-known GitHub organization.
  • References and provides instructions for utilizing official repositories from trusted organizations including Microsoft, Elastic, and SwiftOnSecurity.
  • Downloads supplementary guidance from the Vercel Labs public repository.
  • [COMMAND_EXECUTION]: The skill provides numerous examples and automated scripts for executing commands on the local system for security operations.
  • Includes pipelines for rule conversion using sigma-cli and querying SIEM APIs via Python.
  • Documents the usage of system utilities like autorunsc.exe and sysmon.exe to collect forensic data from local hosts.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 19, 2026, 04:36 PM