The Agent Skills Directory

[COMMAND_EXECUTION]: The publishing CLI tool executes external processes, including system browser launchers (open, xdg-open, or cmd) for authentication flows, and can spawn 'agent-browser' for smoke testing and 'esbuild' for React component bundling.\n- [DATA_EXFILTRATION]: The skill facilitates uploading project fragments and artifact content to a remote API (defaulting to coda0.com). While it supports client-side encryption to protect content, the process involves transmitting project-related data and file hashes to an external server.\n- [DYNAMIC_EXECUTION]: The skill uses the Function constructor to validate the syntax of authored JavaScript fragments at build time. It also dynamically loads the 'esbuild' module for JSX compilation when processing React-format artifacts.\n- [PERSISTENCE]: The tool includes an 'install-hook' command that modifies the agent's environment (e.g., .claude/settings.json) to add a 'Stop' hook. This hook triggers the CLI at the end of turns to automatically detect and flag changes in project files that affect published artifacts.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes external project files as building fragments and allows studying URLs or screenshots for design inspiration. Although it implements explicit procedures to treat remote content as untrusted and uses strict validation to prevent malicious behavior in the final artifact, the ingestion of arbitrary data presents an attack surface.

using-open-artifacts