The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's scripts utilize the subprocess module to call the claude command-line tool. This is a core feature used to execute evaluation tasks and improve skill descriptions.
[EXTERNAL_DOWNLOADS]: The results viewer (viewer.html) fetches the SheetJS library from its official CDN (cdn.sheetjs.com) to render spreadsheet outputs. This download targets a well-known service and is used for functional visualization.
[PROMPT_INJECTION]: The skill processes user-defined prompts and external skill instructions, creating an indirect prompt injection surface.
Ingestion points: Evaluation prompts in evals/evals.json and external instructions from skill directories (e.g., in SKILL.md).
Boundary markers: Isolation is provided by spawning parallel subagents to run test cases separately from the main agent context.
Capability inventory: The tool set includes file system operations (read/write), shell command execution via subprocess, and hosting a local HTTP server.
Sanitization: The skill performs structural and YAML validation on input files but does not sanitize the semantic content of the instructions or prompts.

skill-creator