← 返回 Skills 市场
Model Behavior Layer (Ares MBL)
作者
rushindrasinha
· GitHub ↗
· v1.0.0
· MIT-0
102
总下载
0
收藏
0
当前安装
1
版本数
在 OpenClaw 中安装
/install model-behavior-layer
功能描述
Make any AI model (GPT-5.4, Gemini, Ollama) behave more like Claude. Applies 8 named failure modes reverse-engineered from Claude Code's internal verificatio...
使用说明 (SKILL.md)
Model Behavior Layer (Ares MBL)
What it does
- 8 named failure modes with exact rationalizations and counter-instructions (from Claude Code verificationAgent.ts pattern)
- 3-layer cognitive performance framework (context injection + output rules + critique loop)
- Drop-in system prompt block under 700 tokens for any model
How to use
1. Drop-in system prompt (fastest)
Read MAKE_ANY_MODEL_CLAUDE.md in this skill directory. Copy the "Drop-In System Prompt" block at the end. Paste into your SOUL.md, AGENTS.md, or any system prompt.
2. Full guide
Read MAKE_ANY_MODEL_CLAUDE.md section by section. Apply the failure mode framework progressively.
3. Model-specific
Jump to the Model-Specific Notes section for GPT-5.4, Gemini, or Codex weighting guidance.
When to use this skill
- Switching from Claude to GPT-5.4 or Gemini
- New agent setup (add the drop-in block to SOUL.md)
- Current model is sycophantic, padding responses, or not completing tasks
- Deploying OpenClaw for others who may be on non-Claude models
Source
Reverse-engineered from Claude Code's verificationAgent.ts + Eric (@outsource_) cognitive framework. Built by Ares.
安全使用建议
This is an instruction-only prompt layer that appears coherent with its purpose. Before installing/applying: (1) read the full MAKE_ANY_MODEL_CLAUDE.md and the drop-in prompt so you know exactly what will run as a system prompt; (2) back up your current SOUL.md/AGENTS.md and test the prompt in an isolated/sandbox agent first; (3) if your agent can run shell commands or make network calls autonomously, be cautious — the doc advises running verification commands, and an agent with those privileges could execute them; (4) be aware of any organizational policy about copying or repurposing 'reverse-engineered' internal designs (IP/ethics risk) even though this is not a technical malware indicator. If you want, I can list the exact system-prompt block and highlight any lines you should review before applying.
功能分析
Type: OpenClaw Skill
Name: model-behavior-layer
Version: 1.0.0
The 'model-behavior-layer' skill bundle is a collection of prompt engineering instructions and a system prompt designed to improve AI agent performance by reducing sycophancy, verbosity, and verification avoidance. The content, primarily found in SKILL.md and MAKE_ANY_MODEL_CLAUDE.md, focuses on behavioral alignment and cognitive frameworks to make models like GPT or Gemini behave more like Claude. It contains no executable code, no data exfiltration logic, and no instructions that would lead to unauthorized system access or malicious behavior.
能力标签
能力评估
Purpose & Capability
Name/description match the actual content: an instruction-only prompt layer and behavioral framework. No unexpected binaries, env vars, or installs are requested — everything needed is a system-prompt edit and runtime behavior guidance, which is proportionate to the stated goal.
Instruction Scope
SKILL.md and MAKE_ANY_MODEL_CLAUDE.md instruct the agent to read the included document and copy a drop-in system prompt into SOUL.md/AGENTS.md (expected). The document also encourages running verification commands and 'hitting endpoints' when performing verification — this is guidance to human/agent verifiers and could lead an agent with execution/network privileges to run commands or make network calls. The instructions themselves do not embed any hidden endpoints, credentials, or arbitrary scripts.
Install Mechanism
No install spec and no code files — instruction-only. This minimizes disk-write and supply-chain risk.
Credentials
No environment variables, credentials, or config paths are requested. The documentation asks you to edit your agent's system-prompt files (SOUL.md/AGENTS.md), which is intentional for a prompt-layer but not a secrets request.
Persistence & Privilege
The skill is not forced-always nor requesting platform-level privileges. However, it explicitly asks you to modify your system prompt (SOUL.md/AGENTS.md). Applying those changes affects agent behavior globally — review and test before applying to shared or production agents.
如何使用
- 确保已安装 OpenClaw（本地或 Docker 部署）
- 在对话框中输入安装命令：
/install model-behavior-layer - 安装完成后，直接呼叫该 Skill 的名称或使用
/model-behavior-layer触发 - 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
v1.0.0
Initial release. 8 failure modes from Claude Code verificationAgent.ts, cognitive framework, drop-in system prompt for GPT-5.4/Gemini/Ollama.
元数据
常见问题
Model Behavior Layer (Ares MBL) 是什么？
Make any AI model (GPT-5.4, Gemini, Ollama) behave more like Claude. Applies 8 named failure modes reverse-engineered from Claude Code's internal verificatio... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 102 次。
如何安装 Model Behavior Layer (Ares MBL)？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install model-behavior-layer」即可一键安装，无需额外配置。
Model Behavior Layer (Ares MBL) 是免费的吗？
是的，Model Behavior Layer (Ares MBL) 完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。
Model Behavior Layer (Ares MBL) 支持哪些平台？
Model Behavior Layer (Ares MBL) 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 Model Behavior Layer (Ares MBL)？
由 rushindrasinha（@rushindrasinha）开发并维护，当前版本 v1.0.0。
推荐 Skills