The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to launch local web servers and use headless browser tools like Playwright to create the interactive studio. This execution is scoped to the local development environment and is necessary for the intended functionality.
[DATA_EXFILTRATION]: User-tuned parameters and annotations are exported via structured JSON, URL state, or embedded file metadata to be read back by the agent. This round-trip process is designed for local data persistence and manual ingestion by the agent, rather than exfiltration to external domains.
[PROMPT_INJECTION]: The skill processes user-edited text and comments (ingestion points in SKILL.md) to apply updates to project files (capability). This data is structured as a machine-readable JSON patch (boundary) and processed within a human-in-the-loop workflow, which minimizes the surface for indirect prompt injection.

lookdev