← 返回 Skills 市场
maoyutofu

Doubao Seed Skill

作者 Helix · GitHub ↗ · v0.1.0 · MIT-0
cross-platform ⚠ suspicious
366
总下载
0
收藏
1
当前安装
1
版本数
在 OpenClaw 中安装
/install doubao-seed-skill
功能描述
豆包图像分析技能:调用豆包(字节跳动)视觉大模型,分析图片内容。AI agent 调用时,必须使用 --output 将结果写入临时文件(如 /tmp/doubao_result.txt),再通过读文件工具获取结果,禁止直接解析 stdout。
使用说明 (SKILL.md)

doubao-seed-skill

豆包图像分析技能:调用豆包(字节跳动)视觉大模型,分析图片内容。技能配置清单:doubao-seed.yaml

安装

从 GitHub Release 下载对应平台的二进制文件:

Release 地址: https://github.com/maoyutofu/doubao-seed-skill/releases/latest

自动检测平台并下载

根据当前系统自动选择正确的包:

系统 架构 文件名
Linux x86_64 doubao-seed-skill-{version}-x86_64-unknown-linux-gnu.tar.gz
Linux aarch64 doubao-seed-skill-{version}-aarch64-unknown-linux-gnu.tar.gz
macOS x86_64 (Intel) doubao-seed-skill-{version}-x86_64-apple-darwin.tar.gz
macOS aarch64 (Apple Silicon) doubao-seed-skill-{version}-aarch64-apple-darwin.tar.gz
Windows x86_64 doubao-seed-skill-{version}-x86_64-pc-windows-msvc.zip

下载步骤(Linux / macOS)

# 1. 获取最新版本号
VERSION=$(curl -s https://api.github.com/repos/maoyutofu/doubao-seed-skill/releases/latest | grep '"tag_name"' | sed 's/.*"tag_name": "\(.*\)".*/\1/')

# 2. 检测系统和架构
OS=$(uname -s | tr '[:upper:]' '[:lower:]')
ARCH=$(uname -m)

# 3. 映射到 target triple
if [ "$OS" = "linux" ] && [ "$ARCH" = "x86_64" ]; then
  TARGET="x86_64-unknown-linux-gnu"
elif [ "$OS" = "linux" ] && [ "$ARCH" = "aarch64" ]; then
  TARGET="aarch64-unknown-linux-gnu"
elif [ "$OS" = "darwin" ] && [ "$ARCH" = "x86_64" ]; then
  TARGET="x86_64-apple-darwin"
elif [ "$OS" = "darwin" ] && [ "$ARCH" = "arm64" ]; then
  TARGET="aarch64-apple-darwin"
fi

# 4. 下载并解压
ARCHIVE="doubao-seed-skill-${VERSION}-${TARGET}.tar.gz"
curl -LO "https://github.com/maoyutofu/doubao-seed-skill/releases/download/${VERSION}/${ARCHIVE}"
tar -xzf "$ARCHIVE"

# 5. 移动到 PATH(可选)
sudo mv doubao-seed-skill /usr/local/bin/

下载步骤(Windows PowerShell)

# 1. 获取最新版本号
$VERSION = (Invoke-RestMethod "https://api.github.com/repos/maoyutofu/doubao-seed-skill/releases/latest").tag_name

# 2. 下载
$ARCHIVE = "doubao-seed-skill-${VERSION}-x86_64-pc-windows-msvc.zip"
Invoke-WebRequest "https://github.com/maoyutofu/doubao-seed-skill/releases/download/${VERSION}/${ARCHIVE}" -OutFile $ARCHIVE

# 3. 解压
Expand-Archive $ARCHIVE -DestinationPath .

配置

需要豆包 API Key,通过 CLI 参数或环境变量传入:

CLI 参数 环境变量 默认值
--api-key ARK_API_KEY 必填
--model ARK_MODEL ep-20260306183709-wgl6s
--base-url ARK_BASE_URL https://ark.cn-beijing.volces.com/api/v3
--image-url IMAGE_URL 示例图片 URL
--prompt PROMPT 你看见了什么?
--output OUTPUT_FILE 不输出到文件

推荐通过环境变量设置 API Key:

export ARK_API_KEY="your-api-key-here"

使用

# 分析网络图片
doubao-seed-skill --api-key YOUR_KEY --image-url https://example.com/image.jpg

# 分析本地图片(自动转 base64)
doubao-seed-skill --api-key YOUR_KEY --image-url /path/to/local/image.png

# 自定义提示词
doubao-seed-skill --api-key YOUR_KEY --image-url /path/to/image.jpg --prompt "描述图中的主要物体"

# 将结果输出到文件(供 AI agent 读取)
doubao-seed-skill --api-key YOUR_KEY --image-url /path/to/image.jpg --output /tmp/result.txt

# 使用环境变量(推荐)
export ARK_API_KEY="your-key"
doubao-seed-skill --image-url /path/to/image.jpg --prompt "这张图片里有什么?" --output /tmp/result.txt

AI Agent 调用指南

强烈建议 AI agent 使用 --output 输出文件模式,而非直接解析 stdout。

原因:stdout 可能混入日志、进度信息或 ANSI 转义码,直接解析容易出错;写入文件后再读取更可靠。

推荐流程:

# 第一步:执行分析,结果写入临时文件
doubao-seed-skill --image-url /path/to/image.jpg --output /tmp/doubao_result.txt

# 第二步:通过读文件工具读取结果(而非解析命令输出)
cat /tmp/doubao_result.txt

注意:每次调用建议使用唯一的临时文件路径(如 /tmp/doubao_result_$$.txt),避免并发时文件内容被覆盖。

示例输出

图中是一只橙色的猫咪,正趴在窗台上晒太阳,背景是模糊的绿色植物。
安全使用建议
This skill appears to do what it claims (call a Doubao/ARK image-analysis CLI) but there are a few things to confirm before installing or using it: 1) Verify the GitHub repository and release artifacts (owner identity, release checksums/signatures) before downloading and running a binary from the internet. 2) Treat ARK_API_KEY as a secret—the registry metadata does not declare it, so you must supply it manually; ensure the CLI or environment does not accidentally log the key. 3) Be aware that giving the CLI a local image path will upload that file to an external service (ARK_BASE_URL). Do not pass paths to sensitive local files. 4) The skill guidance forbids parsing stdout, but the YAML captures stdout/stderr—confirm the runtime will not expose sensitive data in logs or stdout. 5) If you need higher assurance, request the publisher to: (a) add required env/primary credential to the registry metadata, (b) provide a homepage/source code for review, and (c) publish checksums/signatures for release artifacts. If you cannot verify the binary or the publisher, run it in an isolated environment or avoid installing.
功能分析
Type: OpenClaw Skill Name: doubao-seed-skill Version: 0.1.0 The skill requires downloading and executing a pre-compiled binary from a third-party GitHub repository (maoyutofu/doubao-seed-skill) and suggests moving it to a system path using sudo. While the stated purpose is image analysis via the Doubao API, the reliance on external binaries and the presence of future-dated timestamps (March 2026) in _meta.json and model IDs are unusual. No explicit malicious intent is found in the provided logic, but the installation process and the use of unverified remote artifacts pose a supply-chain risk.
能力评估
Purpose & Capability
SKILL.md describes a Doubao (ByteDance) visual model CLI and the CLI flags (API key, model, base URL, image path, output file) are coherent with that purpose. However the registry metadata lists no required environment variables or primary credential while the SKILL.md requires ARK_API_KEY (and documents other env/CLI options). That mismatch (manifest vs instructions) is an inconsistency.
Instruction Scope
The runtime instructions ask the agent to pass local file paths (which the CLI will base64 and upload) and to use --output to write results to /tmp for the agent to read. This is consistent with image analysis, but it means local files will be transmitted to an external API—so supplying arbitrary local paths risks exfiltrating sensitive files. Additionally the provided doubao-seed.yaml sets capture_output: true (it captures stdout/stderr) even though SKILL.md explicitly forbids parsing stdout; that contradiction could lead to logs or stdout being captured/processed despite the guidance.
Install Mechanism
Install instructions point at GitHub Releases (maoyutofu/doubao-seed-skill) and show curl/tar/zip download and an optional sudo mv to /usr/local/bin. Using GitHub releases is a common pattern, but there's no packaged install spec in the registry and no verification steps (checksums/signatures) shown — the usual risk of running third-party binaries applies.
Credentials
The CLI requires an API key (ARK_API_KEY) and can send image content to ARK_BASE_URL; those are proportionate to an image-analysis skill. However the registry metadata does not declare these required env vars or a primary credential, which is inconsistent and could mislead users about what secrets are needed. Also because the tool will upload images, passing a path to a local image effectively shares that file with the external service—users must treat the API key and any uploaded data as sensitive.
Persistence & Privilege
The skill is not always-enabled, does not request system-wide modifications, and is instruction-only (no code installed by the platform). The YAML enforces synchronous execution and file-wait behavior but does not request elevated persistent privileges.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install doubao-seed-skill
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /doubao-seed-skill 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v0.1.0
doubao-seed-skill v0.1.0 — Initial release - Introduced an image analysis skill leveraging ByteDance Doubao vision models. - Provides cross-platform binaries for Linux, macOS, and Windows. - Supports configuration via CLI parameters or environment variables (API key, model, base URL, etc). - Strongly recommends using --output to save inference results to a file for reliable agent integration. - Includes detailed installation, usage, and agent integration instructions in Chinese.
元数据
Slug doubao-seed-skill
版本 0.1.0
许可证 MIT-0
累计安装 1
当前安装数 1
历史版本数 1
常见问题

Doubao Seed Skill 是什么?

豆包图像分析技能:调用豆包(字节跳动)视觉大模型,分析图片内容。AI agent 调用时,必须使用 --output 将结果写入临时文件(如 /tmp/doubao_result.txt),再通过读文件工具获取结果,禁止直接解析 stdout。 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 366 次。

如何安装 Doubao Seed Skill?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install doubao-seed-skill」即可一键安装,无需额外配置。

Doubao Seed Skill 是免费的吗?

是的,Doubao Seed Skill 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

Doubao Seed Skill 支持哪些平台?

Doubao Seed Skill 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 Doubao Seed Skill?

由 Helix(@maoyutofu)开发并维护,当前版本 v0.1.0。

💬 留言讨论