AI Agents and Vibe Coding

If you use AI coding agents, this page is for you.

Luthor publishes two docs artifacts specifically for agent context:

• /llms.txt: lightweight index and structured docs map.
• /llms-full.txt: full markdown corpus from all docs pages.

Which file should I load?

• Use /llms.txt when your agent already has repository access and you only need quick docs discovery.
• Use /llms-full.txt when you want the agent to answer package questions directly from one file.
• If token budget allows, load both.

Recommended workflow

1. Load llms-full.txt into your AI agent context.
2. Ask for a task using exact API names, for example:
   • "Create an ExtensiveEditor setup with custom headingOptions and disabled image."
   • "Build a headless editor with createEditorSystem and a minimal toolbar."
3. Ask the agent to point to exact props/methods before coding.
4. Ask the agent to generate code and tests using those exact APIs.
5. Validate by running your app and package tests.

Prompt templates you can reuse

Understand before coding

I loaded Luthor docs in context. Summarize the exact props, defaults, and feature flags needed for this use case:
<your use case>
Then give me a short implementation plan.

Build with exact API usage

Use only documented Luthor APIs from context.
Implement <your task> in my codebase.
Do not invent props or commands.
After changes, list every prop/method you used and why.

Refactor safely

Refactor this editor setup to <target behavior> while preserving output compatibility (JSON/Markdown/HTML).
Show before/after behavior for mode switching and feature flags.

Vibe coding without losing correctness

• Keep prompts concrete: mention preset, mode, feature flags, and output format.
• Ask for command names explicitly (toggleBold, insertTable, setSlashCommands, and so on).
• Ask for fallback behavior (initialMode, defaultEditorView, source conversion errors).
• Ask the agent to verify unsupported combinations before coding.

Practical examples

• "Build a SimpleEditor chat input with submitOnEnter and right-side send button."
• "Create a SlashEditor configuration that only allows heading/list/code block commands."
• "Set up codeHighlightProvider and sync onThemeChange with highlight.js CSS."
• "Use HeadlessEditorPreset and save JSON/Markdown/HTML snapshots from ref methods."

Keep docs and code in sync

When you change behavior:

1. Update docs in apps/web/src/content/docs/**.
2. Run pnpm -C apps/web run sync:docs.
3. Run pnpm -C apps/web run sync:llms.

This keeps agent context aligned with runtime behavior.