# fmode-vision

Claude Code skill for analyzing images and videos via Fmode API vision models (`api.fmode.cn`). Supports single-pass analysis, multi-pass focused analysis with intermediate caching, video-frame analysis, batch processing, and a renovation room-measurement (毛坯房量尺) 5-pass prompt pipeline.

## Install

Project workspace (writes `./.claude/skills/fmode-vision/`):

```bash
npx fmode-vision@latest workspace
```

User level for all workspaces (writes `~/.claude/skills/fmode-vision/`):

```bash
npx fmode-vision@latest install
```

Restart the Claude Code session afterwards so it discovers the skill.

## Token

The skill auto-detects the Fmode API token in this priority order:

1. `FMODE_API_TOKEN` environment variable
2. `~/.fmode/config.json` → `fmodeApiToken` (or `newapiToken`)
3. `<project>/.fmode/config.json` → `fmodeApiToken` (or `newapiToken`)
4. `ANTHROPIC_AUTH_TOKEN` environment variable (zero-config inside Claude Code, since it is the same Fmode gateway token)

If none are found, the skill returns a clear message listing these four options.

## Usage

In Claude Code, just ask in natural language, e.g.:

```
帮我分析这张图片里的关键内容，输出结构化信息。
```

The skill calls `POST https://api.fmode.cn/v1/chat/completions` with the default vision model `doubao-seed-2-0-pro-260215`. Usage is billed against the Fmode token.

## Verify

```bash
npm run smoke
```