free-ai-models
Pass
Audited by Gen Agent Trust Hub on Jul 29, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill incorporates content from external websites into its internal data files and user-facing reports. This creates an indirect prompt injection vulnerability where malicious content on a fetched website could influence the agent's behavior or poison its database.
- Ingestion points: Data retrieved via the
WebFetchtool from external URLs found in reference files or provided by the user. - Boundary markers: There are no instructions to use delimiters or "ignore" markers for the fetched content to prevent potential command leakage.
- Capability inventory: The skill has the ability to write files to the local filesystem, specifically the
reference/andtask/directories. - Sanitization: The skill filters for domain-specific keywords but does not perform security-focused sanitization of the external content before processing or writing it to the local filesystem.
- [EXTERNAL_DOWNLOADS]: The skill is designed to fetch updated information from external sources to ensure data accuracy.
- Evidence: The
WebFetchtool is used to retrieve data from official AI platform websites (e.g., OpenAI, Google, Anthropic, DeepSeek, Alibaba Cloud) for pricing and quota updates. These sources are well-known and reputable official service domains.
Audit Metadata