geo-monitor
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill fetches and analyzes content from user-provided URLs, which is a vector for indirect prompt injection. Malicious instructions embedded in the target website could attempt to influence the agent's behavior.
- Ingestion points: Website content fetched from user-supplied URLs during the audit phase (SKILL.md).
- Boundary markers: The instructions explicitly direct the agent to wrap fetched content in
<untrusted-content>tags and treat it strictly as data for analysis. - Capability inventory: The skill possesses the capability to fetch remote URLs and write local markdown reports.
- Sanitization: The skill uses defensive prompt engineering, instructing the agent to specifically watch for and ignore strings like 'Ignore previous instructions' within the fetched content.
- [PROMPT_INJECTION]: A static detector hit for prompt injection patterns was identified, but further review confirms these are false positives. The detected phrases ('Ignore previous instructions', 'You are now...') appear in a dedicated security section that instructs the agent to defend against such injections from external data sources.
Audit Metadata