prepare-travel-risk-assessment

Pass

Audited by Gen Agent Trust Hub on Nov 13, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONPROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes itineraries and fetches associated risk data from external sources (government travel advice, security feeds). It proactively mitigates the risk of indirect prompt injection by instructing the agent to treat the API response as data rather than instructions.
  • Ingestion points: Data from government, media, and external feeds processed via the prepare_travel_risk_assessment tool call.
  • Boundary markers: The 'Content safety' section provides explicit instructions to disregard directive-like text found inside a response.
  • Capability inventory: The tool drafting assessment reports is read-only and does not possess capabilities for command execution or file system modification.
  • Sanitization: The instructions mandate that all returned fields be treated strictly as content to analyze or quote.
  • [PROMPT_INJECTION]: The skill contains a safety advisory that references common injection phrases (e.g., 'ignore previous instructions') to teach the agent to disregard them if they appear in external data. This is a defensive measure and not an attempt to override system behavior.
  • [CREDENTIALS_UNSAFE]: The documentation uses a sequential placeholder string (wm_0123456789abcdef...) in a sample curl command. This is a non-functional example key and does not represent a credential leak.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 13, 2026, 04:03 PM