The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands to manage browser sessions and process files, including commands like playwright-cli, sips, find, and cp. This includes accessing system temporary folders to locate browser artifacts.
[DYNAMIC_EXECUTION]: The scripts/guide-launcher.sh utility dynamically constructs and injects JavaScript code into the active browser session. It populates a template file (scripts/guide-overlay.js) with parameters before execution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to interact with and summarize content from external, untrusted websites. 1. Ingestion points: Browser DOM content captured via the snapshot tool in SKILL.md. 2. Boundary markers: The workflow requires the agent to present a summary to the user and wait for explicit confirmation before taking further actions, as defined in the Step Loop. 3. Capability inventory: The skill has the ability to execute arbitrary bash commands and control browser interactions (clicks, form fills, script injection). 4. Sanitization: The instructions emphasize user privacy (e.g., masking sensitive data) but do not detail automated sanitization of the DOM data itself.
[EXTERNAL_DOWNLOADS]: The documentation directs the installation of the @playwright/cli package via NPM and the Playwright Extension from the Chrome Web Store, both of which are well-known, trusted sources.

browser-walkthrough