The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to run several bundled scripts using shell and Python interpreters. These include score_skill.py for quality assessment and download_subtitles.sh for fetching YouTube transcriptions via the well-known yt-dlp tool.
[EXTERNAL_DOWNLOADS]: The distillation and fusion pipelines fetch external skill definitions from raw.githubusercontent.com and sitemaps from skills.sh. These operations target established services and official repositories to provide the necessary context and templates for skill building.
[INDIRECT_PROMPT_INJECTION]: As a tool that processes third-party data and skill files, LEAP is exposed to indirect prompt injection vulnerabilities. It attempts to mitigate this through structured parsing logic, mechanical quality filtering (13-point rubric), and the enforcement of specific grammar and boundary rules during the compilation phase.
[DYNAMIC_EXECUTION]: The skill manages a multi-stage pipeline where it creates intermediate JSON and Markdown artifacts, then executes specialized scripts to process these files. This dynamic workflow is constrained to the skill's own directory structure and intended lifecycle.

LEAP