The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct the agent to utilize package runners such as npx, bunx, or pnpm dlx to fetch and execute the latest version of the official clerk CLI tool from the public package registry.
[COMMAND_EXECUTION]: The skill is designed to interact with the host system via the clerk binary to perform authentication, user management, and configuration tasks. It includes specific logic for non-interactive agent modes and detailed exit code handling.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests data from external API responses.
Ingestion points: API responses from the Clerk Backend and Platform APIs are retrieved and stored in temporary files such as /tmp/users.json (SKILL.md, references/recipes.md).
Boundary markers: The instructions explicitly guide the agent to use jq or local scripts (python3, node) to project only necessary data fields, effectively filtering the content before it enters the agent's context.
Capability inventory: The skill uses subprocess calls to the clerk CLI and processing utilities, and performs file operations to update environment variables and temporary data storage.
Sanitization: The skill promotes the use of --dry-run for all mutating operations and utilizes structured JSON output to ensure data consistency and integrity.

clerk-cli