The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill uses a Python script (scripts/download_skill.py) to clone external GitHub repositories to download new agent skills. While many curated sources are provided in references/skill-sources.md, the tool supports downloading from any user-provided URL.
[COMMAND_EXECUTION]: The skill executes system commands via subprocess.run to call git for cloning repositories and managing sparse checkouts within the scripts/download_skill.py file.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and organize external agent instructions (SKILL.md files) from untrusted sources, creating a surface for indirect instruction injection.
Ingestion points: Repository URLs and skill paths processed by scripts/download_skill.py in the root and specified output directories.
Boundary markers: Absent; the system does not wrap downloaded instructions in security delimiters or provide "ignore" directives for embedded content.
Capability inventory: The skill performs directory creation, symlink management, and file writing (shutil.copytree) across various AI tool configuration folders (e.g., .claude/skills, .cursor/skills).
Sanitization: No validation or sanitization of the content within the downloaded skills is performed beyond checking for the existence of a SKILL.md file.

workflow-creator