The Agent Skills Directory

[PROMPT_INJECTION]: The skill explicitly instructs the user to disable security sandboxing using the --sandbox danger-full-access flag, which is a direct request to override the agent's safety boundaries.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface through data ingestion at page.goto(TARGET_URL) in SKILL.md. It lacks boundary markers and sanitization while maintaining high capabilities including subprocess execution and REPL access.
[COMMAND_EXECUTION]: The setup process executes local shell commands such as npm install, npx playwright install, and npm init within the user's workspace.
[REMOTE_CODE_EXECUTION]: Employs dynamic code execution via js_repl for browser automation and launches local Electron binaries using electronLauncher.launch with arbitrary arguments.
[EXTERNAL_DOWNLOADS]: Fetches browser binaries and Node.js packages from public registries during the initialization phase.
[SAFE]: Dependencies and resources are sourced from trusted organizations, specifically Microsoft for the Playwright library.

playwright-interactive