The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes several internal Python scripts provided by the vendor (jurgendn), such as aggregate_benchmark, generate_review.py, run_loop.py, and package_skill.py, to manage the development lifecycle. It also utilizes shell commands like nohup, kill, and open for process management and for opening HTML results in the user's browser.\n- [REMOTE_CODE_EXECUTION]: The skill utilizes a subagent-based testing framework. It spawns independent subagent instances with access to the skill's file path to execute and evaluate test prompts, which is a core feature of the development environment.\n- [DATA_EXFILTRATION]: The skill possesses a data ingestion and export surface through its workflow. It reads user feedback from a feedback.json file generated by a web viewer and exports evaluation sets to the user's ~/Downloads folder. These actions are transparent and necessary for the stated purpose of the tool.\n- [PROMPT_INJECTION]: As a development tool that processes user-defined prompts and evaluation feedback to generate skill instructions, the skill has an inherent surface for indirect prompt injection. Malicious instructions could theoretically be introduced via the test data or user feedback, though the skill includes explicit instructions for the model to avoid creating malicious or misleading content.

skill-creator