thumbnail-machinelisted

# Thumbnail Machine Flow: read the user's content → three rounds of questions → read the format spec + the chosen style template → produce the prompt(s) → delivery gate (MaxFusion MCP / own API key / prompts only). --- ## First run: config check Every time this skill triggers, check whether `config.md` exists in this skill's directory: - **Exists** → read it, go straight to the normal flow. - **Missing** → first use. Run the one-time setup below, then continue. ### One-time setup (ask everything at once) Explain how the skill works first: it turns content into cover-image prompts; generation is optional and happens only at the final delivery gate. Then ask three things: 1. **Default face reference** — covers feature a real person, which needs one clear front-facing photo used as image 1. Two options: - Save one to this skill's directory as `assets/my-face.png` (recommended: auto-used as image 1 on every run) - Choose "upload each time" 2. **Preferred delivery path** — MaxFusion AI MCP (generate in chat), own API key, or prompts only. This becomes the default; the gate still confirms each run. 3. **If prompts-only or own key:** confirm their image model supports **multiple reference images** (image 1 face + image 2+ assets) — e.g. GPT Image, Nano Banana Pro, Seedream 4.0. Single-reference models cannot hold face consistency. Write the answers to `config.md` in this skill's directory (it is gitignored and never committed): # Config - Default face: assets