geo-monitoring-setup
Pass
Audited by Gen Agent Trust Hub on Nov 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests external brand context and query lists (e.g., queries.txt) which are interpolated into prompts sent to external AI APIs. While this is inherent to the task, it creates a surface where malicious input could influence the behavior of the targeted models.
- Ingestion points: File 'queries.txt' and user-provided brand context description.
- Boundary markers: Absent; the script sends queries verbatim without delimiters or instructions to ignore embedded commands.
- Capability inventory: Network requests to OpenAI, Anthropic, Google, and Perplexity APIs; local file write to 'geo-sample-*.json'.
- Sanitization: Absent; input is passed directly to API call methods.
- [EXTERNAL_DOWNLOADS]: The script template references standard, well-known libraries (openai, anthropic, google-generativeai, requests) for interacting with official AI services. These are established dependencies for the intended functionality.
Audit Metadata