The Agent Skills Directory

[COMMAND_EXECUTION]: The run.js script facilitates the execution of arbitrary Node.js code by writing input to a temporary file in the skill directory and loading it via the require() function.
[REMOTE_CODE_EXECUTION]: The skill architecture is based on the agent dynamically writing Playwright scripts to the /tmp directory and executing them via a wrapper, allowing for the execution of arbitrary logic.
[DATA_EXFILTRATION]: The detectDevServers helper in lib/helpers.js performs local network discovery by probing common development ports (3000, 3001, 3002, 5173, 8080, 8000, 4200, 5000, 9000, 1234) on localhost.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data from external websites. Ingestion points: Data extracted using extractTexts and extractTableData in lib/helpers.js. Boundary markers: No delimiters or isolation warnings are used in the instructions or code. Capability inventory: Full Node.js execution and file system access. Sanitization: No validation or filtering is applied to web content before it enters the agent's context.
[EXTERNAL_DOWNLOADS]: The setup script in package.json and the run.js executor download the Playwright library and Chromium browser binaries from the official NPM registry and Playwright's distribution servers.

playwright-skill