local-ai-agents
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [Indirect Prompt Injection Surface]: The skill provides instructions for building agents that process external data via documentation search (RAG) and file tools. This creates an interface where untrusted data could potentially influence the agent's instructions. The skill proactively addresses this by recommending that all tool operations be scoped to a sandboxed project directory and by encouraging output validation, which aligns with security best practices.
- [Local Command Execution]: The instructions involve the use of the
foundryCLI and local tool execution. The skill emphasizes that tools run with the user's permissions and must be sandboxed. This is a standard pattern for developer workstation tools and is presented here within the context of local development and privacy-preserving AI design. - [Credential Management]: The code samples demonstrate the use of an API key variable (
manager.api_key) described as a local placeholder for compatibility with the OpenAI protocol. This is a safe practice for local development that avoids hardcoding actual secrets.
Audit Metadata