The Agent Skills Directory

COMMAND_EXECUTION (HIGH): The run.js script takes arbitrary input from process.argv or stdin and writes it to a temporary file (.temp-execution-[timestamp].js), which is then executed using require(). This allows the execution of any Node.js code on the host system.
EXTERNAL_DOWNLOADS (MEDIUM): The installPlaywright function automatically invokes npm install and npx playwright install if the Playwright dependency is missing, performing network operations and package installation at runtime without user confirmation.
REMOTE_CODE_EXECUTION (HIGH): The execution flow allows untrusted input to be converted directly into executable code, providing a high-risk surface for Remote Code Execution (RCE) via the agent's input channel.
PROMPT_INJECTION (LOW): The skill is susceptible to indirect prompt injection as it lacks sanitization for the code it executes.
Ingestion points: run.js (reads from process.argv and process.stdin).
Boundary markers: Absent; no delimiters or instructions to ignore embedded commands are present.
Capability inventory: run.js contains fs.writeFileSync, require() (execution), and execSync (package installation).
Sanitization: Absent; the input is wrapped in a basic template and executed without validation.

playwright-skill