agentic-ai-developer
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill consists of markdown guides and architectural patterns for building AI agents. It does not contain any executable code, hardcoded credentials, or suspicious network operations. It emphasizes defensive security practices.\n- [INDIRECT_PROMPT_INJECTION]: The skill addresses the processing of external, untrusted data by AI agents.\n
- Ingestion points: The documentation (specifically
references/security_and_production_deployment.md) identifies user messages, RAG chunks, web fetches, email bodies, and tool return payloads as untrusted input channels.\n - Boundary markers: The skill recommends using delimiters like
<untrusted>...</untrusted>and instructions to ignore embedded commands in the system prompt.\n - Capability inventory: The skill guides the implementation of tools for searching, database queries, and code execution as described in
references/agent_loop_tools_and_mcp.md.\n - Sanitization: It advocates for schema validation, stripping HTML/scripts, and normalizing URLs to mitigate injection risks.
Audit Metadata