dt-sec-ioc-hunting

Pass

Audited by Gen Agent Trust Hub on Aug 14, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface in ioc-intake.md by processing external threat reports and advisory URLs. 1. Ingestion points: Untrusted text from pasted reports and fetched advisory URLs. 2. Boundary markers: The skill uses structured DQL templates to isolate IoC values, though explicit markers for the raw input text are not defined in the query body. 3. Capability inventory: Execution of DQL queries on logs and spans. 4. Sanitization: ioc-intake.md includes logic to defang URLs (e.g., replacing hxxp), strip STIX wrappers, and validate IP address formats.
  • [COMMAND_EXECUTION]: The skill dynamically generates domain-specific language (DQL) queries by interpolating extracted IoCs into templates found in hunt-logs.md and hunt-spans.md. While standard for this workflow, the assembly of queries from external inputs requires strict adherence to templates to prevent injection.
  • [SAFE]: The skill implements strong resource management and safety controls: hunt-logs.md enforces IoC chunking to prevent query complexity failures; timeframe-gating.md mandates user approval for search expansion beyond 15 minutes; and secondary-observable-extraction.md provides a recursion guard to limit automatic re-hunting to one level.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 14, 2026, 03:45 PM
Security Audit — agent-trust-hub — dt-sec-ioc-hunting