Meitu Beauty

对人像照片进行 AI 美颜处理（磨皮、美白、精修五官）。当用户提到美颜、磨皮、美白、精修、beautify、beauty enhance、让照片更好看时触发。仅支持单人照片。

Security Scan

The skill's requirements and runtime instructions are coherent with an image-beautification CLI wrapper: it needs the meitu CLI and Meitu API credentials, and it reads/writes only the declared local workspace and credentials file.

Purpose & Capability

Name/description (AI beautify single-person portrait) match the declared requirements: a 'meitu' CLI binary and Meitu AK/SK. Declared file reads/writes (~/.meitu/credentials.json and the visual workspace) align with storing credentials and output images.

Instruction Scope

SKILL.md stays focused on: validating a single-person photo, calling `meitu image-beauty-enhance`, handling outputs and retries. It requires downloading/saving user-provided images (local path, URL, or uploaded image) and will upload them to the Meitu service via the CLI — this is expected but important for privacy. One minor ambiguity: the doc uses '目视检查' (visual check) for face detection; it's not explicit whether the agent uses a local face-detector or relies solely on the meitu CLI. That affects whether local face-detection code would run, but nothing in the skill claims to read unrelated files.

Install Mechanism

Instruction-only skill (no install spec). It suggests installing meitu-cli via `npm install -g meitu-cli`, which is a reasonable, proportional dependency for wrapping that CLI. No downloads from arbitrary URLs or archive extraction are declared.

Credentials

Only MEITU_OPENAPI_ACCESS_KEY and MEITU_OPENAPI_SECRET_KEY are required (with AK as primary). These directly map to the Meitu API use described. The skill also declares reading ~/.meitu/credentials.json as an alternate credential source — consistent with the credential setup instructions.

Persistence & Privilege

always:false and no system-wide modifications. The skill reads/writes only its declared workspace path and the credentials file; it does not request elevated system privileges or modify other skills. Autonomous invocation is allowed (platform default) but not combined with other high-risk factors.

Assessment

This skill appears to do what it says: it calls the meitu CLI and requires your Meitu AK/SK, and will read/write files in ~/.openclaw/workspace/visual/ and ~/.meitu/credentials.json. Before installing, consider: (1) you will need to install a third‑party npm CLI (meitu-cli) and provide API credentials; (2) images processed will be uploaded to Meitu's service via that CLI — avoid sending sensitive/private photos if you do not want them transmitted; (3) confirm you trust the meitu-cli package source (npm) and the Meitu service; (4) if you want the agent to pre-validate faces locally, ask the skill author to clarify whether local face-detection is used or if validation relies solely on Meitu's responses. No regex-scan warnings were present, and nothing in the skill requests disproportionate credentials or unrelated system access.

Overview

一键 AI 美颜：磨皮、美白、精修五官。调用 meitu image-beauty-enhance，仅支持单人人像照片。

Dependencies

• meitu-cli ≥ 0.1.9 — npm install -g meitu-cli
• 凭证配置: meitu config set-ak --value "..." + meitu config set-sk --value "..." 或环境变量 MEITU_OPENAPI_ACCESS_KEY / MEITU_OPENAPI_SECRET_KEY

路径别名： 下文中 $VISUAL = {OPENCLAW_HOME}/workspace/visual/

Core Workflow

Preflight → [Context: 跳过（工具型美颜，无创意自由度）] → Execute → Deliver

Preflight

1. meitu --version → 未安装则提示 npm install -g meitu-cli
2. meitu auth verify --json → 凭证无效则引导配置
3. Detect mode: cwd has openclaw.yaml → project mode; else → one-off 检查 $VISUAL 目录 → 确定 capabilities
4. output_dir 解析（Preflight 内 MUST 完成）： Resolve output_dir: openclaw.yaml → ./output/ | else → $VISUAL/output/meitu-beauty/ mkdir -p {output_dir}

Execute

输入获取

用户提供图片，接受以下形式：

• 本地文件路径 → 使用 --image <path>
• 图片 URL → 使用 --image <url>
• 对话中直接发送的图片 → 保存到临时文件后使用 --image <path>

若用户未提供图片，主动询问："请提供需要美颜的人像照片（文件路径或 URL）。"

人像校验（关键前置步骤）

此工具仅支持单人人像。调用前 MUST 校验：

1. 读取用户提供的图片，目视检查：

   • 是否包含人脸 → 无人脸则拒绝："这张图片中没有检测到人像，美颜工具仅支持包含人脸的照片。"
   • 是否为单人 → 多人则拒绝："检测到多人，美颜工具仅支持单人照片。请裁剪为单人后重试。"
   • 人脸是否足够大且清晰 → 人脸过小/模糊则警告："人脸较小/模糊，美颜效果可能不明显，是否继续？"

2. 若无法预判（如 URL 无法预览），直接调用工具，根据返回错误处理。

强度选择

用户意图	--beatify_type	说明
"自然美颜"、"轻微调整"、"稍微美化"、未指定强度	0	自然效果（默认）
"大力美颜"、"重度磨皮"、"效果强一点"、"狠狠美颜"	1	增强效果

调用命令

meitu image-beauty-enhance \
  --image <url_or_path> \
  --beatify_type <0|1> \
  --json \
  --download-dir {output_dir}

注意: 参数拼写是 --beatify_type（非 --beautify_type），这是 CLI 的已知拼写。

结果处理

• ok: true → --download-dir 已指定，从 downloaded_files[0].saved_path 获取本地文件路径；若未使用 --download-dir，从 media_urls[0] 获取结果图片 URL
• ok: false → 进入错误降级

错误降级

先检查 error_name 分流，再按级别降级：

error_name	处理	可重试
CONTENT_REQUIREMENTS_UNMET (code 98501)	图片不含人脸或不符合单人要求。直接告知用户："该图片不符合美颜要求（需要单人人像照片），请更换图片。" 不重试。	否
ORDER_REQUIRED	余额不足，提示充值，展示 action_url	否
CREDENTIALS_MISSING	提示配置 AK/SK	否
其他错误	按以下级别降级 ↓	视情况

通用降级（仅 retryable: true 或未知错误时）：

级别	策略	操作
L1	降低强度	beatify_type 1 → 0 重试
L2	检查图片质量	提示用户更换更清晰、人脸更大的照片
L3	检查图片格式	确认为 JPG/PNG/WEBP，非 GIF/BMP 等不支持格式
L4	停止报错	连续 2 次失败 → 报告 code + hint，停止重试

Deliver

直接使用 Preflight 解析的 output_dir。

mv {file} {output_dir}/{date}_beauty_{original_name}.{ext}

命名示例：2026-03-23_beauty_portrait.jpg

Output

项目	规格
格式	与原图一致（通常 JPG）
数量	单张
命名	{date}_beauty_{original_name}.{ext}
位置	由 Deliver 步骤解析