skills/different-ai/agent-bank/ai-sdk/Gen Agent Trust Hub

ai-sdk

Pass

Audited by Gen Agent Trust Hub on Jul 15, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Uses curl and jq to retrieve and process a live list of available AI models from the Vercel AI Gateway (ai-gateway.vercel.sh). This allows the agent to identify and use the most recent model versions.
  • [EXTERNAL_DOWNLOADS]: Recommends installing the official ai npm package using standard package managers (e.g., pnpm add ai) and fetches documentation results from ai-sdk.dev. These resources originate from trusted organizations.
  • [PROMPT_INJECTION]: The skill instructs the agent to disregard its internal training data in favor of information retrieved at runtime from local files and official remote APIs. This identifies a surface for indirect prompt injection, although the risk is mitigated by the use of trusted sources.
    • Ingestion points: node_modules/ai/docs/, ai-sdk.dev/api/search-docs.
    • Boundary markers: None observed.
    • Capability inventory: Shell execution (curl, grep, jq), package installation (pnpm).
    • Sanitization: Not explicitly implemented; the agent relies on the integrity of the official SDK documentation.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 15, 2026, 02:25 AM
Security Audit — agent-trust-hub — ai-sdk