agent-designer
Pass
Audited by Gen Agent Trust Hub on Sep 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The toolkit processes user-provided JSON files (containing system requirements, tool descriptions, and execution logs) to generate structured output intended for consumption by AI agents.
- Ingestion points: The scripts
agent_planner.py,tool_schema_generator.py, andagent_evaluator.pyread external JSON files that contain natural language strings. - Boundary markers: The generated schemas and Mermaid diagrams do not wrap user-supplied descriptions in boundary markers or include instructions to ignore embedded commands, potentially allowing descriptive text to influence downstream agent behavior.
- Capability inventory: The toolkit scripts are restricted to local data processing using standard Python libraries. They lack dangerous capabilities such as network exfiltration, arbitrary command execution (subprocess), or dynamic code evaluation (eval/exec).
- Sanitization: There is no evidence of sanitization or filtering applied to the input text to detect or mitigate prompt injection patterns that could be present in user-supplied descriptions.
Audit Metadata