The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents the use of CLI tools via npx for application scaffolding (create-polpo-app) and component installation (@polpo-ai/ui add).
[EXTERNAL_DOWNLOADS]: Instructions include the installation of several Node.js packages from the public npm registry, including @polpo-ai/chat, @polpo-ai/sdk, @polpo-ai/react, react-virtuoso, lucide-react, and streamdown.
[DYNAMIC_EXECUTION]: The theming reference includes a recommended anti-flash script for dark mode that uses dangerouslySetInnerHTML to execute a localized JavaScript function that reads from localStorage and modifies the document's class list.
[INDIRECT_PROMPT_INJECTION]: The skill builds interfaces that render agent-generated messages and execute client-side tools, creating a surface for indirect prompt injection.
Ingestion points: Data ingested through the useChat hook and the <Chat> component, specifically the messages array and tool call definitions.
Boundary markers: The documentation does not mandate specific delimiters for isolating agent content from system instructions within the UI.
Capability inventory: The framework supports client-side tool execution (continueToolResult), durable SSE connections, and authenticated file URL resolution.
Sanitization: The skill promotes safety by instructing developers to validate client tool names and arguments against a local registry before execution.

polpo-react