The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses shell commands (ls, grep, rg, test) to inspect the local filesystem for agents and configuration scripts located within the ~/.claude/ directory.
[DYNAMIC_EXECUTION]: Inline JavaScript is executed via bun -e to perform complex logic, such as validating symlinks, parsing YAML frontmatter from agent files, and performing atomic updates to the settings.json file.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from local agent files which could theoretically be manipulated to influence the agent's reporting.
Ingestion points: Reads content from ~/.claude/skills/workflows/ and ~/.claude/agents/*.md.
Boundary markers: Uses programmatic parsing in Node.js with regex to target specific keys (skills:) rather than processing full text as instructions.
Capability inventory: The skill has file read capabilities, filesystem enumeration, and targeted write access to ~/.claude/settings.json.
Sanitization: Employs JSON.parse and JSON.stringify to ensure configuration integrity, though it relies on standard regex for metadata extraction from markdown files.

setup