The Agent Skills Directory

[COMMAND_EXECUTION]: The agent is instructed to run the validation script validate.py using uv run python. This is a standard part of the workflow to ensure data integrity.
[DYNAMIC_EXECUTION]: Skill scripts (validate.py, generate_reference.py, coverage.py) dynamically modify the Python search path (sys.path) to import local project modules from the services/ directory. This is used to leverage existing project validation logic.
[INDIRECT_PROMPT_INJECTION]: The skill processes workflow JSON files containing variable references (e.g., {{webhook.*}}). This represents a theoretical surface for indirect prompt injection if untrusted data were passed through a workflow. * Ingestion points: The validate.py script reads workflow JSON from the local file system. * Boundary markers: Relies on json.loads for structural parsing. * Capability inventory: Local script execution and access to internal project modules. * Sanitization: Structural validation of node data keys and variable reference checking.

flow-builder