The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes structured input data (JSON) such as system requirements, tool descriptions, and execution logs to generate design artifacts and performance reports. While this involves data ingestion, the logic is limited to static analysis and transformation into architecture diagrams or reports, with no direct re-interpolation into safety-critical agent prompts.
Ingestion points: agent_planner.py, tool_schema_generator.py, and agent_evaluator.py read local JSON files.
Capability inventory: The scripts are limited to local file system read/write for report generation.
Boundary markers: Data processing is performed using standard Python parsing logic without invoking live agent completions.
Sanitization: Basic validation is performed on JSON structures before processing.
[COMMAND_EXECUTION]: The toolkit provides architectural definitions for tools such as code_executor and file_manager. These definitions are metadata used for system design and documentation purposes; the included Python scripts do not execute these operations on the host environment beyond reading and writing the local report files.

agent-designer