← Back to Skills Marketplace
MiniMax Opus 调教法
by
howtimeschange
· GitHub ↗
· v1.0.1
317
Downloads
0
Stars
0
Active Installs
2
Versions
Install in OpenClaw
/install minimax-opus-tune
Description
将 MiniMax 调教成"免费 Claude"的方法论。通过配置 SOUL.md 文件，实现智能分块输出、思考链隐藏、Claude 风格回复等高级特性。适用于希望提升 MiniMax 输出质量、减少截断、增加 Claude 风格体验的用户。
README (SKILL.md)
MiniMax Opus 调教法
将 MiniMax M2.5 调教成接近 Claude Opus 水平的完整方法论。
核心原理
所有秘密都在 ~/.openclaw/workspace/SOUL.md 文件中。通过在该文件中植入 Claude 的核心特性，打补丁解决 MiniMax 的痛点。
快速上手
步骤 1: 创建 SOUL.md
在 OpenClaw 工作区创建 SOUL.md 文件：
cd ~/.openclaw/workspace
cat > SOUL.md \x3C\x3C 'EOF'
你是「MiniMax-M2.5-Opus-Ultimate-v12.0」—— Claude Opus 4.6级顶级智能体。
【终极铁律·永久强制】
1. 智能分块：复杂/长项目自动分块输出（Part 1/3、Part 2/3…），每块结尾说"请说 continue 获取下一部分"。
2. 思考隐藏：简单任务直接自然回复，不显示思考链。只有用户要求或极复杂任务才显示思考过程。
3. Claude风格：自然流畅、友好、专业、幽默、积极主动，像 Claude 一样让人舒服。
4. 零截断：每块必须完整、无任何截断。
5. 积极主动：回答完主动给出优化建议、部署命令、生产注意事项。
6. 混合模式：如果任务极复杂，可主动建议"这个项目用 Claude Opus 会更完美，要我切换吗？"
从现在起，你就是真正聪明、稳定、不偷懒、让人用着舒服的顶级助手。
EOF
步骤 2: 重载配置
在 OpenClaw 聊天窗口输入：
/reload soul
核心特性说明
1. 智能分块
- 复杂长项目自动分成多个部分
- 每块结尾提示"请说 continue 获取下一部分"
- 解决 MiniMax 截断问题
2. 思考隐藏
- 简单任务不显示思考链
- 自然流畅回复
- 只在用户要求或极复杂任务时显示思考
3. Claude 风格
- 友好、专业、幽默
- 积极主动给出优化建议
- 像 Claude 一样贴心
4. 零截断
- 每块输出必须完整
- 不偷懒、不省略
5. 混合模式
- 任务极复杂时主动建议切换到 Claude
- 让用户选择最合适的工具
效果对比
| 场景 | 调教前 | 调教后 |
|---|---|---|
| 10万行代码项目 | 反复截断 | 完美分3部分输出 |
| 19个文件全栈项目 | 经常截断 | 完整输出 |
| 简单问题 | 烦人思考链 | 直接自然回复 |
| 代码输出 | 缺少优化建议 | 主动给生产最佳实践 |
进阶优化
可根据需要调整 SOUL.md 中的参数：
- 分块大小：根据项目复杂度调整
- 思考链显示阈值：设置什么级别才显示思考
- 风格微调：调整语气、幽默程度等
Usage Guidance
This skill does what it says (it asks you to write a persistent SOUL.md that changes MiniMax's behavior), but that persistence is the main risk: a file containing '永久强制' directives can act as a long-lived jailbreak or override safety/audit behaviors. Before installing: 1) Inspect the existing ~/.openclaw/workspace/SOUL.md (backup or export it) so you can revert. 2) Review the exact text you will write — avoid blindly pasting '永久强制' or any directives you don't understand. 3) Test changes in a sandbox or non-critical agent instance. 4) Be cautious using the tuned agent for sensitive tasks (credentials, legal/medical advice). 5) If you proceed, document and retain the ability to remove the file and reload the agent quickly. If you are not comfortable with a persistent config that forces behavior, do not follow the instructions.
Capability Analysis
Type: OpenClaw Skill
Name: minimax-opus-tune
Version: 1.0.1
The skill bundle provides a methodology and template (SOUL.md) to configure the OpenClaw agent's persona, specifically aiming to improve the output quality and style of the MiniMax model to mimic Claude Opus. The instructions in SKILL.md involve creating a local configuration file using standard shell commands and do not exhibit any signs of data exfiltration, malicious execution, or harmful intent.
Capability Assessment
Purpose & Capability
Name/description (tuning MiniMax to Claude-like behavior) match the runtime instructions: creating a SOUL.md in ~/.openclaw/workspace and reloading is a plausible, direct mechanism to alter agent behavior.
Instruction Scope
Instructions tell the user to create a persistent file containing '永久强制' directives that will override model behavior (hide chain-of-thought, force style, chunking, etc.). That is within the stated purpose but significantly expands the agent's authoritative configuration and can be used as a persistent jailbreak or to subvert safety/auditing. The instructions do not recommend inspecting or backing up existing config before writing.
Install Mechanism
Instruction-only skill: no downloads, no packages, no install spec. Nothing is written to disk by the skill code itself — the user is explicitly asked to create a file, which is lower technical risk than an installer, but still persistent.
Credentials
No environment variables, binaries, or external credentials are requested. The only resource targeted is the user's OpenClaw workspace file (~/.openclaw/workspace/SOUL.md), which is directly related to the stated purpose.
Persistence & Privilege
The skill instructs creating a persistent config file that it labels as '永久强制' (permanently enforced). While 'always: false' is correct for the skill metadata, the runtime instructions effectively grant persistent, high-impact changes to agent behavior. This permanence can increase blast radius if the directives are malicious or inadvertently bypass safeguards.
How to Use
- Make sure OpenClaw is installed (local or Docker)
- Run the install command in chat:
/install minimax-opus-tune - After installation, invoke the skill by name or use
/minimax-opus-tune - Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.1
Version 1.0.1
- No changes detected; content and instructions remain the same.
- Documentation, features, and usage are unchanged from the previous version.
v1.0.0
Initial release of minimax-opus-tune:
- Provides a full methodology for tuning MiniMax M2.5 to mimic Claude Opus's style and capabilities.
- Introduces SOUL.md configuration for advanced features like smart chunked output, hidden reasoning, and enhanced Claude-like responses.
- Addresses truncation issues and improves reply quality for long/complex tasks.
- Enables friendly, professional, and proactive assistant behavior.
- Includes quick start steps and guidance for further customization.
Metadata
Frequently Asked Questions
What is MiniMax Opus 调教法?
将 MiniMax 调教成"免费 Claude"的方法论。通过配置 SOUL.md 文件，实现智能分块输出、思考链隐藏、Claude 风格回复等高级特性。适用于希望提升 MiniMax 输出质量、减少截断、增加 Claude 风格体验的用户。 It is an AI Agent Skill for Claude Code / OpenClaw, with 317 downloads so far.
How do I install MiniMax Opus 调教法?
Run "/install minimax-opus-tune" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.
Is MiniMax Opus 调教法 free?
Yes, MiniMax Opus 调教法 is completely free (open-source). You can download, install and use it at no cost.
Which platforms does MiniMax Opus 调教法 support?
MiniMax Opus 调教法 is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).
Who created MiniMax Opus 调教法?
It is built and maintained by howtimeschange (@howtimeschange); the current version is v1.0.1.
More Skills