Agent workflow

The whole loop lives in files and commands

Tileflow is designed so an AI coding agent can make map changes safely: generate a config, run validation, preview through the app, inspect the result, and deploy with an API key from the environment.

The local edit–validate–preview loop needs no account.Only the hosted deploy step requires approved access and a project key.

Small typed surfacePrefer fields like basemap, theme, and view instead of hand-writing hundreds of raw MapLibre layers.

Deterministic feedbacktileflow validate turns config mistakes into actionable errors that an agent can fix in the next edit.

Terminal

# 1. Edit the map as code
$EDITOR tileflow.config.ts
 
# 2. Ask Tileflow for deterministic feedback
tileflow validate
 
# 3. Preview through your app dev server
pnpm dev
 
# 4. Deploy when the preview is right
tileflow deploy

Starter rule file

Put a short agent instruction file in app starters so Claude Code, Cursor, and similar tools learn the Tileflow workflow as soon as they open the repo.

CLAUDE.md

# CLAUDE.md / .cursorrules
- Style maps in tileflow.config.ts.
- Prefer basemap, themes, and view over raw MapLibre layer JSON.
- After every config change, run: tileflow validate.
- Preview with the app dev server before deploy.
- Use named maps such as "madrid" from React: <Map map="madrid" />.