The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The run.js script implements a mechanism to execute arbitrary JavaScript code at runtime. It takes input from command-line arguments or stdin, writes it to a temporary file in the skill's directory, and executes it using the Node.js require() function. This allows for the execution of any code generated by the agent or provided via user input.\n- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it extracts untrusted information from web pages.\n
Ingestion points: Data enters the agent context via the extractTexts and extractTableData functions in lib/helpers.js, which evaluate and return text from the active browser page.\n
Boundary markers: There are no instructions or implementations of boundary markers or delimiters to isolate extracted web content from the agent's instructions.\n
Capability inventory: The skill possesses high-privilege capabilities, including the ability to execute arbitrary JavaScript (run.js), perform network operations, and execute shell commands (npm install and playwright install via execSync in run.js).\n
Sanitization: The helper functions in lib/helpers.js do not perform any sanitization, filtering, or validation of the text content retrieved from websites before passing it to the agent.\n- [COMMAND_EXECUTION]: The run.js script and package.json file contain logic to execute shell commands using execSync and npm scripts, primarily for environment setup and installing browser binaries.\n- [EXTERNAL_DOWNLOADS]: During setup, the skill downloads Playwright from the NPM registry and browser binaries from Microsoft's official infrastructure. These are documented as fetches from well-known and trusted sources.

playwright-skill