The Agent Skills Directory

[Environment Restrictions Reduction]: The skill requires users to disable standard security filters by instructing them to run the environment with the configuration parameter --sandbox danger-full-access (or sandbox_mode=danger-full-access). While stated as a temporary requirement for Playwright support, this step removes typical execution constraints and exposes the host platform to any subsequent commands executed by the agent.
[External Tools & Module Installation]: The setup script executes commands (npm install playwright and npx playwright install chromium) that download third-party libraries and pre-compiled binaries at runtime. This introduces potential supply chain risk if dependency resolution configurations are compromised, though the packages are standard for UI testing automation.
[Dynamic Code Execution via Evaluator]: The skill leverages a persistent JavaScript REPL environment (js_repl) and uses dynamic import() to load modules locally. It also heavily relies on executing scripts via browser contexts (page.evaluate and electronApp.evaluate). If any of the data processed within these web pages originates from untrusted sources, it exposes the agent workflow to context-level inputs.
[Indirect Prompt Injection Surface]: The skill actively crawls, automates, and reads state information from local web applications or external web paths via page.goto(). This establishes an ingestion point for external data into the agent context without explicitly enforced boundary markers or input sanitization mechanisms, making it susceptible to unexpected interactions if the target page contains adversarial content.

playwright-interactive