The Agent Skills Directory

Safe Script Execution Environment: The skill guides the agent to generate and run Playwright scripts (final_script.py) locally using Firefox in a headless configuration (playwright.firefox.launch(headless=True)). Artifacts like screenshots and log files are stored under a local workspace directory (final_runs/run_<id>/). No unexpected external execution vectors or data storage surfaces are used.
Absence of Remote Downloads: The guidelines explicitly instruct the agent not to install extra packages via pip or apt (Do not install extra packages with pip/apt), and no unverified external software components are referenced.
No Unsafe Credentials: The configuration eliminates the requirement for model keys or external verification services (replaces the OpenAI-backed image_qa and self_reflection tools with your own native abilities... No OPENAI_API_KEY or other model API keys required), mitigating credential exposure risks.
Standard Tool Utilization: Access to tools like Bash, Read, Write, and Edit aligns with the defined manifest structure and matches the primary intent of code generation, localized exploration, and verification tasks.

webwright