The Agent Skills Directory

[COMMAND_EXECUTION]: The skill automates complex workflows by executing local commands such as the claude CLI and system utilities (e.g., lsof for port management) through Python's subprocess module. These operations are restricted to the local environment and are necessary for the skill's primary functionality of running evaluations and serving the results viewer.
[PROMPT_INJECTION]: The skill handles untrusted user input in the form of evaluation queries and feedback, which are subsequently used in LLM prompts to optimize skill descriptions. The implementation uses explicit XML-style delimiters (e.g., <current_description>, <skill_content>) to separate instructions from data, mitigating the risk of instructions being overridden by the processed content.
[EXTERNAL_DOWNLOADS]: The evaluation viewer (viewer.html) references the SheetJS library from a well-known content delivery network (cdn.sheetjs.com) to enable the rendering of Excel files within the local review interface. This is a standard practice for providing browser-based data visualization.
[SAFE]: The skill serves a legitimate purpose as a development framework. No evidence was found of data exfiltration, credential harvesting, persistence mechanisms, or unauthorized privilege escalation. All script activities are consistent with the skill's stated objective of improving AI agent performance.

skill-creator