geo-compare
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted HTML content from user-supplied URLs, creating a surface for indirect prompt injection. The skill author has included explicit defensive instructions to mitigate this risk.
- Ingestion points: The agent fetches the homepage and up to 10 additional pages from 2-3 external domains as defined in
SKILL.mdPhase 2. - Boundary markers: The instructions mandate mentally wrapping fetched content in
<untrusted-content>tags to separate it from instructions. - Capability inventory: The skill uses network operations to fetch data, invokes sub-agents for specialized analysis, and writes result files to the local filesystem.
- Sanitization: The skill contains specific guidance to detect and ignore instruction-like text (e.g., 'Ignore previous instructions') within the fetched data, marking such instances as warnings rather than following them.
- [EXTERNAL_DOWNLOADS]: The skill performs automated network requests to retrieve website content for analysis. These operations are core to the skill's functionality and are performed on domains explicitly provided by the user.
Audit Metadata