Work & OperationsDone with custom-built or unnamed agent Feb 19, 2026
Let agents fill out web forms using text-grid page rendering
Author built TextWeb, a text-grid browser so LLM agents can fill out job applications without costly vision models.
- What They Did
- The author built TextWeb, which loads a page in headless Chromium and pulls out each visible element's position, text, and whether it's interactive. It maps these onto a character grid that keeps the on-screen layout. Interactive elements get numbers, so an agent issues commands like click 3 or type 7 followed by text. Any LLM can read it.
- How It Went
- Each page comes out as roughly 2-5KB of text instead of a roughly 1MB screenshot, so no vision model is needed. The post gives no results from actual job applications and asks for feedback on the grid format.
- Worth Knowing
- It installs with npm install -g textweb and plugs in through an MCP server, function-calling definitions, LangChain, CrewAI, an HTTP API, a CLI, or a Node.js library.
Try it yourself with custom-built or unnamed agent
Write a small tool that renders a web page as a text grid, listing each form field, button, and link with a reference number, so an LLM agent can fill out forms without screenshots. Test it on [job application or form URL] using placeholder data and show me the filled fields. Never submit the form.
Do this with a skill.
These SkillShop playbooks do this exact job.