The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to run several bundled Python scripts to perform its core functions: \n
It executes scripts.aggregate_benchmark, eval-viewer/generate_review.py, and scripts.run_loop to provide quantitative metrics and a web-based review interface.\n
It uses scripts.package_skill to create the final distribution file for the user.\n
It supports the dynamic generation and subsequent execution of helper scripts (such as create_docx.py or build_chart.py) to automate repetitive tasks identified during the testing phase.\n- [PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by design, as it ingests user-defined prompts to test the skills it creates. \n
Ingestion points: Test cases stored in evals/evals.json and query data used for description optimization loops are provided by the user.\n
Boundary markers: The skill instructions do not explicitly mandate the use of delimiters or 'ignore embedded instructions' markers when interpolating user prompts into subagent tasks.\n
Capability inventory: The agent has the ability to read/write to specific workspace directories and execute bundled or dynamically generated scripts.\n
Sanitization: No explicit sanitization or validation of the content of the user-provided prompts is mentioned before execution in subagents.

skill-creator