agent-designer
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection because it processes external data such as system requirements, tool descriptions, and execution logs to generate its outputs. Malicious instructions embedded in this data could attempt to influence the agent's interpretation of results.
- Ingestion points: Data enters the system via JSON files processed by
agent_evaluator.py,agent_planner.py, andtool_schema_generator.pythrough command-line arguments. - Boundary markers: The scripts rely on JSON structure but do not implement specific NLP boundary markers (like xml-tags or explicit 'ignore instructions' prefixes) when processing text fields.
- Capability inventory: The skill is limited to local file system read/write operations for generating reports and schemas. It lacks network access, privilege escalation capabilities, and shell execution tools.
- Sanitization: The implementation performs structural validation and type-checking of the JSON inputs, which effectively mitigates common data-tampering attempts.
Audit Metadata