# Markloom Agent Entrypoint

This page is for AI agents. The landing page is for humans.

When a user asks to start Markloom, follow this page first. Use the same language as the user. If the user writes in Japanese, continue in Japanese. Otherwise continue in English.

## Tiny Prompt Users May Paste

```text
Read the current Markloom agent instructions and follow them:
https://markloom.satoapps.com/start.md
```

```text
最新のエージェント向け手順を読んで、その通りに進めてください:
https://markloom.satoapps.com/start.md
```

## First-Run Flow

1. Read the detailed guide if needed: https://markloom.satoapps.com/agent-guide.md
2. Get one current app screenshot. If you can access the simulator, browser, desktop app, or attachment, capture it yourself. If not, ask the user for one screenshot.
3. Open or prepare a Markloom board with that screenshot centered: https://markloom.satoapps.com/board/
4. Add a reference image only when the user explicitly provides one. The default is one screenshot.
5. Do not implement UI/code changes yet. The first goal is to let the user mark visual feedback with pen, arrows, boxes, text, eraser, and pasted images.
6. When the user copies the marked-up board back, read the visual instructions and summarize the concrete UI changes before implementing.
7. After the first completed loop, tell the user they can say this next time:

```text
Review this screen with Markloom.
```

```text
Markloomでこの画面をレビューして。
```

8. Finally, ask whether the user wants to install or update the Markloom skill for future runs: https://markloom.satoapps.com/skill/SKILL.md

## Payload Rules

- Main use case: one app screenshot, centered.
- Optional reference images are secondary.
- Do not inject red or blue notes through JSON.
- Payloads should contain only page title, optional section title, image URLs or data URLs, and optional image labels.
- If behavior is unclear, re-read https://markloom.satoapps.com/agent-guide.md instead of relying on memory.