agent-vision-diagrams
Pass
Audited by Gen Agent Trust Hub on Jul 18, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill instructs the agent to fetch a markdown file from a user-provided URL and 'follow it' as the single source of truth for the session. This creates a surface for indirect prompt injection where the external content could command the agent to perform unauthorized actions. Ingestion points: Content from
<app-url>/skill.md. Boundary markers: Absent; the agent is explicitly told to follow the remote instructions. Capability inventory: The agent can perform HTTP requests, utilize authentication tokens, and execute diagramming tools. Sanitization: Absent; no validation is performed on the fetched content. - [EXTERNAL_DOWNLOADS]: The skill performs network requests to arbitrary URLs provided by the user or found in data to fetch instructions and rendering outputs, leading to interactions with unverified infrastructure.
- [DATA_EXFILTRATION]: The skill handles authentication tokens via environment variables and bearer headers. A risk exists where an attacker could provide a malicious URL to trick the agent into sending these tokens to an external server.
Audit Metadata