The Agent Skills Directory

Unverifiable Dependencies & Remote Code Execution (HIGH): The skill recommends executing code from untrusted remote sources. Specifically, it instructs the agent to run npx @tailwindcss/cli and npx @tailwindcss/upgrade in the references/build-cli-package.md and references/build-node-version.md files. Since the @tailwindcss organization is not on the whitelisted Trusted External Sources, this constitutes a download-and-execute pattern from an unverified source.\n- Indirect Prompt Injection (HIGH): This skill has a high vulnerability surface for indirect injection. \n
Ingestion points: The agent is directed to read and analyze project-specific CSS, TSX, and JSX files.\n
Boundary markers: None provided in the instructions to prevent the agent from obeying instructions embedded in user-controlled code.\n
Capability inventory: The skill encourages the use of npx and pnpm shell commands, providing a bridge from data processing to system execution.\n
Sanitization: No validation or sanitization of the code being processed is mentioned.\n- Command Execution (MEDIUM): The README.md and references/build-cli-package.md files encourage the use of build and validation scripts (pnpm install, pnpm build, pnpm validate). While standard for developers, for an autonomous agent, this grants significant control over the local environment for non-whitelisted operations.

tailwind