8917 Minimax Toolkit

MiniMax 多模态工具集。用于图片生成、图生图、视频生成、视频模板、语音合成、长文本 TTS、声音克隆、声音设计与音乐生成。适用于需要调用 MiniMax 官方 API 处理文本、图片、音频或视频素材的场景。默认将产物输出到当前工作区的 `workspace/03-Resources/minimax-outp...

Security Scan

The skill's code, instructions, and requested MINIMAX_API_KEY are coherent with a MiniMax API wrapper for multimodal generation; no evidence of hidden exfiltration or unrelated credential requests was found.

Purpose & Capability

Name/description, required MINIMAX_API_KEY, CLI entrypoints, and scripts all align with a MiniMax multimodal API wrapper for image/video/audio generation. The files and references describe model IDs, endpoints, quota logic and output locations that match the stated purpose.

Instruction Scope

Runtime instructions and scripts explicitly read MINIMAX_API_KEY (or fallback to ~/.openclaw/openclaw.json), upload user-supplied media to MiniMax endpoints, and save outputs into workspace or configured output dirs. This is within the stated scope, but it means user media and the Token Plan key will be sent to third-party MiniMax endpoints — the skill discloses this in SKILL.md and references. No instructions attempt to read unrelated system files or secrets.

Install Mechanism

No install spec provided (instruction-only/CLI scripts). All code is present in the bundle and there is no remote download/install step. This is the lower-risk pattern for skills.

Credentials

The skill only requires MINIMAX_API_KEY (declared). The code also optionally reads ~/.openclaw/openclaw.json (declared in the docs) as a fallback and reads MINIMAX_OUTPUT_DIR to pick output location — MINIMAX_OUTPUT_DIR is used but not listed in the declared requires.env. No unrelated credentials (AWS, GitHub, etc.) are requested.

Persistence & Privilege

Skill does not request always:true, does not modify other skills or global agent settings, and only writes outputs to user/workspace directories. Autonomous invocation is enabled by default (normal) but not combined with elevated privileges or broad credential access.

Assessment

This skill appears to be what it claims: a MiniMax API wrapper that will upload any media you give it and will read a MiniMax Token Plan key from MINIMAX_API_KEY or ~/.openclaw/openclaw.json. Before installing or running it: (1) confirm the key you provide is the correct Token Plan key and that you accept third‑party uploads; (2) avoid sending sensitive images/audio unless you consent to MiniMax processing; (3) be aware outputs are saved under workspace/03-Resources/minimax-output/ (or ./outputs/minimax/ or MINIMAX_OUTPUT_DIR if set); (4) consider running in an isolated environment if you are in a high-security setting; (5) note that feature_flags.json disables some heavy modalities (video/music) by default — review that file if you expect those capabilities. The only small inconsistency: the code honors MINIMAX_OUTPUT_DIR even though it isn't listed in the skill's declared required env vars. Otherwise no hidden endpoints, obfuscated code, or unrelated credential requests were found.

定位

这是一个 工具型 / Tool Wrapper skill。

用它在 MiniMax 官方 API 上执行多模态生成任务，并把产物稳定落到当前工作区或用户指定目录。

核心原则

1. 执行前先提示消耗：先输出 Token Plan 请求消耗预估，再执行。
2. 执行前先讲清隐私边界：涉及图片、音频、视频等素材上传时，明确告知会发送到 MiniMax API。
3. 优先遵守输出纪律：优先写入当前工作区；没有 workspace/ 时再退回通用输出目录。
4. 项目隔离优先：需要长期管理时优先使用 --project。
5. 禁止硬编码 API Key：仅从 MINIMAX_API_KEY 或 ~/.openclaw/openclaw.json 读取。

输出纪律

输出根目录优先级：

1. --output-dir
2. MINIMAX_OUTPUT_DIR
3. 当前工作目录下的 workspace/03-Resources/minimax-output/
4. 当前目录不存在 workspace/ 时，退回 ./outputs/minimax/

使用 --project <ProjectName> 时，产物进入项目子目录。

执行完成后，始终明确告诉用户：

• 文件类型
• 保存路径
• 如需长期管理，建议整理到项目目录

预算与隐私

执行前至少说明：

• 模型名称
• 预计消耗的 request 次数
• 5 小时滚动窗口规则
• 高消耗任务风险提示（尤其视频）

涉及私密图片、音频或视频时，先确认用户接受第三方 API 处理。

详细说明见：

• references/budget-and-trust.md
• references/api_info.md
• references/costs.json

能力导航

按具体场景读取：

• references/modalities.md：查看 9 类能力、对应脚本和 CLI 示例
• references/budget-and-trust.md：查看预算、上传与隐私边界
• references/api_info.md：查看当前模型 ID 与基础约束
• references/quota_mapping.json：查看运行时模型名与官方额度桶的映射
• references/official-doc-sources.md：查看需要定期核验的官方来源
• references/troubleshooting.md：查看联网失败、网页抓取失败、remains 查询失败时的排障指引

官方文档校验机制

可运行：

python3 scripts/mm.py remains
python3 scripts/mm.py check-docs

其中：

• mm.py remains：查询官方 Token Plan 实时额度
• mm.py check-docs：抓取官方 FAQ + 查询 remains + 对比本地 references

校验脚本会：

1. 抓取官方 Token Plan FAQ
2. 查询官方 coding_plan/remains 接口
3. 对比本地 references/ 中的关键配置
4. 生成差异报告到 references/checks/latest-check.md

如果在线校验失败：

• 不阻塞整个 skill
• 退回本地 references 离线模式
• 输出排障指引
• 详见 references/troubleshooting.md

推荐使用统一入口

优先使用统一入口 scripts/mm.py：

# Image / Video / Music
python3 scripts/mm.py image "A red apple" --ratio 16:9 --project Demo
python3 scripts/mm.py video "Cinematic ruins" --project Demo
python3 scripts/mm.py music "Lo-fi beat" --instrumental --project Demo

# Speech (short text)
python3 scripts/mm.py speech "Hello world" --voice male-qn-qingse --project Demo

# Speech (long text / file)
python3 scripts/mm.py async-speech ~/script.txt --voice male-qn-qingse --project Demo

# Image-to-image
python3 scripts/mm.py i2i "anime style" --ref ~/photo.jpg --project Demo

# Video templates
python3 scripts/mm.py video-template labubu --media ~/photo.jpg --project Demo

# Voice clone / design
python3 scripts/mm.py voice-clone ~/my_voice.wav --voice-id my-voice
python3 scripts/mm.py voice-design "Warm deep male voice" --voice-id new-voice

支持的 modality：image, i2i, video, video-template, speech, async-speech, voice-clone, voice-design, music

当前支持显式功能开关：references/feature_flags.json

• 可按 modality 设定 true / false
• 当前套餐不支持的能力建议直接设为 false，避免无意义调用

注意：

• voice-design 当前要求 --preview-text
• voice-clone 建议提供 --preview-text，如提供 --prompt-audio 则需同时提供 --prompt-text
• 复刻音色若 7 天内未正式调用，官方会删除该音色

详细示例见 references/modalities.md。

兼容入口（旧脚本仍可用）

如需直接调用或处理更复杂场景，以下脚本仍可用：

• scripts/mm_image.py
• scripts/mm_i2i.py
• scripts/mm_video.py
• scripts/mm_video_template.py
• scripts/mm_speech.py
• scripts/mm_async_speech.py
• scripts/mm_voice_clone.py
• scripts/mm_voice_design.py
• scripts/mm_music.py

旧脚本作为兼容层保留，未来逐步收敛到统一入口。

一句话原则

MiniMax 产物默认落当前工作区或通用输出目录，执行前讲清 request 消耗与第三方 API 处理边界。