ai-sdk
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEPROMPT_INJECTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- Functional override of internal knowledge: The skill instructions direct the agent to disregard potentially outdated internal training data in favor of current documentation. This is a common pattern for technical assistants to ensure accurate code generation and is used here for grounding purposes.\n- Retrieval from official sources: The skill fetches model information and documentation from Vercel-owned domains like vercel.sh and vercel.ai. These operations are used to provide the latest API details and model versions directly from the vendor.\n- Indirect prompt injection surface:\n
- Ingestion points: The agent is directed to read local files in node_modules/ai/docs/ and query ai-sdk.dev for documentation (SKILL.md).\n
- Boundary markers: The instructions do not specify explicit delimiters to isolate the content of fetched documentation from the agent's system instructions.\n
- Capability inventory: The skill utilizes command-line tools such as curl, jq, and grep, as well as Node.js package managers to manage the environment.\n
- Sanitization: No explicit sanitization or filtering of the documentation content is performed before processing.\n- Development Tooling: The skill references DevTools and codemod packages (@ai-sdk/devtools, @ai-sdk/codemod) which can be executed via npx for debugging and migration, which is standard development workflow behavior.
Audit Metadata