The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill employs a custom modular loading system (load_skill, load_ref) to dynamically retrieve and display markdown-based instructions and reference files. These modules are sourced from computed local paths identified through extensive environmental probing of common skill installation directories (e.g., ~/.claude/skills, ~/.agents/skills, /sessions, /mnt/skills/plugins).
[INDIRECT_PROMPT_INJECTION]: The pipeline includes a 'HARVEST' phase that ingests content from external sources, including client social profiles, existing websites, and provided documents. This untrusted data is synthesized by agents into a PRODUCT.md file used to direct subsequent site-building activities, creating an entry point for indirect prompt injection attacks.
[EXTERNAL_DOWNLOADS]: The skill documentation and error handling logic reference external resources for installation and updates, specifically suggesting the use of npx to fetch code from vendor-controlled domains and repositories (genjutsu.athevon.dev, github.com/athevon/genjutsu).
[COMMAND_EXECUTION]: The skill orchestrates a wide range of shell operations across parallel clones, including the execution of rsync, find, node, headless Chrome, and framework-specific binaries. It also executes a Python-based design audit script (audit.py) retrieved from the dynamically resolved skill base directory.

bunshin