The Agent Skills Directory

[COMMAND_EXECUTION]: The skill employs a dynamic execution model where browser automation scripts are generated at runtime and executed via a wrapper.
Evidence: The run.js script writes provided code to a temporary file and loads it using the Node.js require() function. It also utilizes execSync to manage its own installation and setup environment.
[EXTERNAL_DOWNLOADS]: The skill manages the lifecycle of its automation engine by downloading necessary packages and browser binaries from well-known registries.
Evidence: The setup script and run.js perform npm install and npx playwright install to acquire the Playwright framework and the Chromium browser.
[DATA_EXFILTRATION]: The skill performs local network operations to facilitate connection to development environments.
Evidence: The detectDevServers helper in lib/helpers.js probes a predefined list of common ports on localhost to identify active web servers for automated testing.
[REMOTE_CODE_EXECUTION]: The skill has an inherent exposure to indirect prompt injection due to its core function of processing untrusted web content.
Ingestion points: Data is ingested from external web pages using Playwright locators, table extraction helpers, and browser console logs.
Boundary markers: No explicit delimiters or isolation instructions are present in the automation scripts to distinguish between intended instructions and untrusted data returned from the browser.
Capability inventory: The agent possesses the capability to write JavaScript files to the filesystem and execute them via a Node.js wrapper.
Sanitization: Page content is extracted and reported to the agent context without specialized sanitization or escaping mechanisms.

playwright-skill