The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes grep for code indexing and playwright-cli for browser interaction. The commands are structured and purposeful, following standard automation patterns for UI testing and inspection.
[EXTERNAL_DOWNLOADS]: The instructions suggest installing @playwright/cli from the official npm registry if it's missing. This is a well-known, industry-standard package for browser automation.
[REMOTE_CODE_EXECUTION]: The skill uses playwright-cli eval to execute JavaScript snippets within the browser context. The provided scripts are benign, focusing on extracting bounding boxes, computed styles, and accessibility attributes from the DOM.
[DATA_EXFILTRATION]: While the skill captures screenshots and DOM content, this data is used locally by the agent to answer user queries. There is no evidence of data being transmitted to unauthorized remote servers.
[INDIRECT_PROMPT_INJECTION]: The skill represents a surface for indirect prompt injection because it ingests untrusted data from web pages (DOM snapshots). However, the skill explicitly advises truncating the outerHTML to approximately 600 characters, which serves as a mitigation factor for token limits and reduces the likelihood of processing large, malicious payloads hidden in page content.

locate-ui-from-code