The Agent Skills Directory

COMMAND_EXECUTION (MEDIUM): The skill executes local scripts such as ./scripts/dev-runner.sh and Playwright CLI commands. While intended for environment setup, executing arbitrary local scripts carries inherent risk if the script content is not verified.
EXTERNAL_DOWNLOADS (MEDIUM): The skill triggers npm install and npx playwright install chromium, which download and install external packages and browser binaries. Without a lockfile or pinned versions specified in the skill, this introduces a dependency on external registries.
DYNAMIC_EXECUTION (MEDIUM): The core workflow involves the 'Generator' phase, which writes executable TypeScript files (.spec.ts). This code is dynamically created by an AI agent and subsequently intended for execution, which is a significant security surface.
INDIRECT_PROMPT_INJECTION (LOW):
Ingestion points: The 'Planner' agent reads data from a running web application's UI to generate test specifications.
Boundary markers: No specific delimiters or safety instructions are mentioned to separate untrusted UI data from the test generation logic.
Capability inventory: The workflow includes the ability to write files to the filesystem and execute them via the Playwright runner.
Sanitization: There is no evidence of sanitization for the data scraped from the UI before it is used to generate executable code.

playwright-e2e