The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests user input through the $ARGUMENTS placeholder in SKILL.md to define dataset characteristics. This input directly influences the generation of Python scripts and CLI arguments, creating a surface for prompt injection.
Ingestion points: $ARGUMENTS in SKILL.md is used to describe the synthetic dataset to be built.
Boundary markers: No explicit delimiters or boundary markers are used to isolate the user-provided description from the generation instructions.
Capability inventory: Writing Python files (customer_reviews.py, etc.), executing CLI commands (data-designer), and running local utility scripts (scripts/get_person_object_schema.py).
Sanitization: No sanitization or validation of the user input is implemented before it is processed.
[COMMAND_EXECUTION]: The workflows in workflows/autopilot.md and workflows/interactive.md instruct the agent to execute the data-designer CLI for tasks like validation, previewing, and creation. It also requires running a local script scripts/get_person_object_schema.py to inspect managed persona datasets.
[PRIVILEGE_ESCALATION]: In the Troubleshooting section of SKILL.md, the agent is instructed to suggest disabling sandbox isolation to resolve network errors. While this requires user permission, recommending a security bypass to the user reduces the overall safety posture of the execution environment.
[EXTERNAL_DOWNLOADS]: The SKILL.md output template utilizes PEP 723 inline metadata to declare dependencies such as data-designer and pydantic. These packages are resolved and downloaded from external registries during script execution.

data-designer