The Agent Skills Directory

[DATA_EXFILTRATION]: The setup agent in agents/clerk-setup.md is instructed to search for and read sensitive Clerk authentication keys from .env files and configuration files using grep and find.
[COMMAND_EXECUTION]: The skill includes executable logic in scripts/generate-session-token.js and shell workflows in the setup agent for managing authentication tokens and environment secrets, which involves executing raw commands with high-privilege access.
[METADATA_POISONING]: The SKILL.md file contains false information about a 'March 2025 CVE' and Next.js 16 requirements to justify non-standard project structures, which can mislead the agent's security reasoning.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of user-controlled data via Clerk webhooks and custom JWT claims without establishing boundary markers or sanitization protocols to prevent the agent from executing embedded malicious instructions.
[EXTERNAL_DOWNLOADS]: The skill automates the installation of packages from NPM and interacts with remote API endpoints; while these targets are well-known services, the autonomous execution of these downloads creates a significant attack surface.

clerk-auth