Webwright is a browser agent framework that enables language models to complete web tasks by writing and executing Python code that controls browser sessions.
The tool addresses the limitation of step-by-step web interaction loops by inverting the typical agent-browser relationship. Rather than having a model predict one action at a time within a predefined interaction harness, Webwright gives the model a terminal environment where it can launch multiple browser sessions, inspect page states through screenshots when needed, and write executable Python code. The persistent artifact is the code script itself, not the browser session history. This approach treats the browser as a tool the agent can invoke and discard while developing a program, enabling the model to reason at a higher level of abstraction.
Webwright suits projects where web task automation needs to be reproducible and debuggable as code. The framework works with OpenAI, Anthropic, and OpenRouter backends, and uses Playwright for browser control. It can be integrated as a plugin or skill into existing agent systems like Claude Code, Codex, Hermes, and OpenClaw, making it useful for teams that already have preferred coding models and want to extend their capabilities to browser tasks. The tool's minimal footprint and terminal-based approach make it accessible for developers who prefer working with code rather than complex multi-agent orchestration systems.
The project ships with a Skill Factory feature that distills solved tasks into reusable, parameterized code skills that can run standalone without invoking the model. The team has added support for Task2UI mode, which renders task results into HTML-based web applications for easy viewing and reuse. Integration manifests for multiple agent platforms have been released, allowing the same skills folder to load across different systems. The codebase maintains a small footprint while supporting multiple LLM backends and browser automation through Playwright.
