A Claude Code markdown workflow that never leaves one window

The best Claude Code markdown workflow is the one where you never touch Alt-Tab. You write the spec in markdown, run the agent against it, read the diff it produces, and jot notes for the next prompt — and every one of those steps happens in the same window. That is exactly the loop we built Tybre.md around: a WYSIWYG markdown editor with a real terminal and an inline browser preview living side by side.

In our own daily use, the friction was never Claude Code itself. It was the tab-dance around it: spec in one editor, agent in a terminal app, output copied into a third window, notes in a fourth. This post walks through the loop end to end with the actual commands we run.

The Alt-Tab tax on AI-assisted coding

Claude Code is a terminal agent. You point it at files, it reads them, plans, and edits. The moment your spec lives in a separate app, you are shuttling context by hand — copying a paragraph out, pasting a path in, flipping back to check the result. Each hop is small, but the loop runs dozens of times an hour.

Keeping the spec, the terminal, and the preview in one surface removes that tax. The agent reads the same file you are editing; the output prints where you can see it; your notes sit under the spec. If you want the deeper argument for why the files should stay on your disk instead of a cloud silo, we made it in our local-first notes writeup.

The loop: draft prompt, run agent, review, update notes

Here is the exact sequence. Steps 1 and 5 happen in the editor pane; steps 2 to 4 in the terminal pane; the browser pane is there when the agent spins up a dev server.

• —Draft the spec. Open a new .md file and write your PRD or task spec in the WYSIWYG editor — headings, checklists, and code fences render as you type, so it reads like the finished document.
• —Start the agent. Open the built-in terminal in the same window, cd into the project if needed, and run claude to launch Claude Code in that directory.
• —Point it at your spec. Prompt it plainly — implement the plan in spec.md — and Claude Code reads the very file you just wrote. No copy-paste.
• —Review the diff. Claude Code prints its plan and per-file edits in the terminal; run git diff to inspect changes, and when it starts a dev server the inline browser preview shows the result beside the code.
• —Update notes in place. Below the spec, jot what worked, what to retry, and the next prompt. The document becomes a running log of the session.
• —Repeat. Refine the prompt, re-run, re-review. The loop stays in one window from draft to done.

Real commands you run in the built-in terminal

The terminal is a full PTY (xterm.js over portable-pty), so anything that works in your shell works here. A typical session mixes agent and plumbing commands:

• —claude — start Claude Code in the project; add a prompt inline or type it interactively.
• —git diff and git add -p — review and stage exactly what the agent changed before you trust it.
• —npm run dev — the agent (or you) starts the dev server, and the inline browser preview opens the localhost URL.
• —npm run type-check and npm run lint — verify the agent's edits actually compile before the next round.

Why one window beats three

Multiple sessions per project mean you can run Claude Code in one terminal tab and keep a shell free for git in another, without leaving the app. Instant Ctrl+1 to Ctrl+9 project switching and session restore mean the whole layout — open spec, terminal history, preview URL — comes back exactly as you left it after a restart.

None of this locks up your files. Everything is plain markdown on disk, no cloud account required. If you are weighing Tybre.md against a pure writing tool, our comparison with Typora is honest about where a dedicated WYSIWYG editor still wins — and where a built-in terminal changes the equation. For the wider field, see the best markdown editors for developers in 2026.