The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently executes the claude CLI via subprocess in scripts/run_eval.py and scripts/improve_description.py to perform triggering evaluations and description optimization. Additionally, eval-viewer/generate_review.py executes system commands like lsof to manage local network ports for its internal review server.
[INDIRECT_PROMPT_INJECTION]: The skill implements an evaluation loop that processes user-defined test prompts stored in evals/evals.json. These prompts are executed as tasks via subagents or the CLI without explicit sanitization or strict boundary enforcement, creating a surface where malicious instructions within test data could potentially influence the agent's behavior during the evaluation process.
[EXTERNAL_DOWNLOADS]: The eval-viewer/viewer.html template fetches the SheetJS (xlsx.full.min.js) library from the cdn.sheetjs.com CDN to enable browser-based previewing of spreadsheet outputs. This is a well-known service used for the intended purpose of the viewer.

skill-creator