The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands to discover existing configurations and validate new agents. Evidence: Phase 1 utilizes ls and cat to inspect agents in the .claude/agents/ directory; Phase 7 and 8 utilize uvx skilllint and claude plugin validate for system validation.
[EXTERNAL_DOWNLOADS]: The skill employs uvx to fetch and run the skilllint package from the Python Package Index (PyPI) to ensure agent frontmatter compliance. PyPI is a standard and well-known service for this environment.
[PROMPT_INJECTION]: The skill displays an indirect prompt injection surface as it reads and adapts content from existing local agent files. Ingestion points: Reads all files in .claude/agents/*.md and ~/.claude/agents/*.md for context gathering and adaptation. Boundary markers: Content is read without specific boundary delimiters or instructions to disregard embedded instructions. Capability inventory: The skill has permissions to write/edit files, execute shell commands, and manage subagents. Sanitization: No input sanitization or logic validation is performed on the content of the adapted files.

agent-creator