The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes playwright-cli, npx, and npm to execute browser automation scripts and manage the runtime environment. These tools are invoked with various arguments to drive user interactions and system tests.
[EXTERNAL_DOWNLOADS]: During initialization, the skill fetches the @playwright/cli package and browser binaries (Chromium, Firefox, WebKit) from official NPM and Playwright repositories. These are well-known, trusted services used for development purposes.
[REMOTE_CODE_EXECUTION]: Includes the run-code and eval commands, which allow the agent to execute arbitrary JavaScript within the browser context. This functionality is a core requirement for advanced automation tasks like geolocation spoofing and complex DOM manipulation.
[CREDENTIALS_UNSAFE]: Provides mechanisms to capture and restore browser states, including cookies and localStorage, using commands like state-save and cookie-get. The documentation includes a dedicated security section advising users to avoid committing resulting .json files to version control.
[DATA_EXFILTRATION]: Possesses the capability to read sensitive data from web pages (via snapshots and eval) and write it to the local filesystem (via screenshots and state files). This is intended for legitimate testing and auditing purposes.
[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection as it processes content from external, potentially untrusted web pages. However, it is primarily a tool for development environments and includes no instructions that attempt to bypass safety filters.

playwright-cli