The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (HIGH): The skill promotes the use of npx shadcn@latest add https://ui.inference.sh/r/tools.json. This command fetches and executes code from an unverified remote source (ui.inference.sh), which poses a supply chain risk.
[REMOTE_CODE_EXECUTION] (HIGH): The instruction to run npx skills add inference-sh/skills@agent-ui uses a custom package system (skills) and remote packages (inference-sh/skills) that are not part of the trusted ecosystem, allowing for arbitrary code execution on the user's machine.
[COMMAND_EXECUTION] (HIGH): The documentation contains multiple instances of shell commands (npx) intended to be run by the user. These commands perform network operations and local file modifications.
[PROMPT_INJECTION] (HIGH): (Category 8: Indirect Prompt Injection surface) The ToolCall, ToolResult, and ToolApproval components (SKILL.md) ingest untrusted data via args and result props. There are no visible boundary markers or sanitization logic in the provided React components. The skill handles tool lifecycle events, including human-in-the-loop approvals (onApprove) and execution of side-effect-heavy tools like send_email or read_file, creating a high-severity capability tier.

tools-ui