The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the Python subprocess module across several scripts (e.g., run_eval.py, package_skill.py) to execute the claude CLI and system utilities like lsof and kill for managing evaluation runs and local viewer instances.- [EXTERNAL_DOWNLOADS]: The evaluation viewer template (viewer.html) loads the SheetJS (xlsx) library from the cdn.sheetjs.com content delivery network. This well-known service is used to render spreadsheet files generated during evaluations within the browser.- [DATA_EXFILTRATION]: As part of its optimization loop, improve_description.py transmits the content of the SKILL.md file and evaluation metrics to the Anthropic API. This data transmission is the core intended functionality for generating improved skill metadata through a trusted vendor service.- [PROMPT_INJECTION]: The skill manages an environment for running user-provided test prompts through subagents. This creates an indirect prompt injection surface where the skill creator processes potentially untrusted data from evaluation query files (evals.json) and passes it into an executable context.

skill-creator