← 返回 Skills 市场
459
总下载
0
收藏
2
当前安装
1
版本数
在 OpenClaw 中安装
/install world-model
功能描述
World Model - Environment understanding, causal reasoning, and prediction for AGI
安全使用建议
This package is incomplete and internally inconsistent. Before installing or enabling it: 1) Ask the author for the missing runtime files (world_model.py and the referenced world-model-api.ps1) and for provenance (source repository and release signatures). 2) If you receive those files, inspect them for network calls, subprocess execution, file I/O, and credential use (search for requests, socket, subprocess, os.system, open, exec, importlib.exec_module). 3) Do not run in production or grant broad permissions until you verify behavior; test inside a restricted sandbox and monitor outbound connections. 4) Prefer skills that declare required env vars, config paths, and explicit install steps — large capability claims without corresponding code or declared permissions are a red flag. If you want, provide the missing world_model.py and any install scripts and I can re-evaluate with higher confidence.
功能分析
Type: OpenClaw Skill
Name: world-model
Version: 2.0.0
The `unified_wrapper.py` file passes the `goal` parameter, which originates from the AI agent's prompt, directly to methods of the underlying `world_model.py` skill (e.g., `_original.do(goal)`). This design introduces a significant prompt injection vulnerability, as there is no apparent sanitization or validation of the `goal` string within the wrapper. If the unanalyzed `world_model.py` module were to interpret this input as executable code or commands, it could lead to remote code execution. This is a critical vulnerability that allows for potential attacks, rather than evidence of intentional malicious behavior by the skill itself.
能力评估
Purpose & Capability
The SKILL.md promises an AGI-level world model that monitors 50+ system variables, simulates actions, and can 'try actions before executing', but the package declares no required binaries, env vars, or config paths. The shipped files are static JSON data and a wrapper; the actual implementation (references to world-model-api.ps1 and an expected world_model.py) is missing. This mismatch suggests the package is incomplete or intentionally missing runtime code.
Instruction Scope
The documentation includes PowerShell API examples (world-model-api.ps1) and API functions that imply reading/updating world state and performing simulations and risk assessments. Those examples would normally require scripts and code to interact with the host system or other tools, but those files are not present. The SKILL.md also describes capabilities like anomaly detection and 'simulate before acting' which imply capability to run commands or invoke other tools; this scope is broader than the declared package content and permissions.
Install Mechanism
No install spec (instruction-only) which is low-risk. However the included unified_wrapper.py performs a dynamic import of a local world_model.py (absent from the bundle). If a runtime world_model.py were present, that dynamic loading would execute arbitrary Python code from the skill directory — expected for many skill wrappers but worth reviewing in any complete package. Currently nothing is installed, but the wrapper's behavior means a future or alternative package containing world_model.py could run arbitrary code.
Credentials
The skill requests no environment variables or credentials, yet claims to monitor system/network/tools and to execute/simulate actions. world-state.json lists tools including 'exec' and 'browser' (implying command execution or external access). The absence of declared env/credential requirements is not proportionate to the stated capabilities and increases uncertainty about what the real implementation would require or access.
Persistence & Privilege
The skill is not marked always:true and declares no config paths or persistent privileges. disable-model-invocation is false (normal default), so the agent could invoke the skill autonomously — this is expected platform behavior and not by itself a flag. Combine this with other concerns before allowing autonomous use.
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 world-model - 安装完成后，直接呼叫该 Skill 的名称或使用
/world-model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
v2.0.0
World Model Skill v2.0.0
- Major update providing comprehensive environment modeling, causal reasoning, and predictive analytics for AGI applications.
- Adds real-time state tracking (50+ variables), anomaly detection, and causal chain analysis with up to 5 levels depth.
- Introduces robust prediction and simulation functions, supporting what-if analysis, risk assessment, and outcome forecasting.
- Includes detailed PowerShell API for state management, causal inference, prediction, and simulation actions.
- Benchmarked performance: 85% prediction accuracy, 92% causal reasoning, simulations under 50ms.
元数据
常见问题
World Model 是什么？
World Model - Environment understanding, causal reasoning, and prediction for AGI. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 459 次。
如何安装 World Model？
在 OpenClaw 或 Claude Code 对话框中运行命令「/install world-model」即可一键安装，无需额外配置。
World Model 是免费的吗？
是的，World Model 完全免费（开源免费），可自由下载、安装和使用。
World Model 支持哪些平台？
World Model 跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（darwin, linux, win32）。
谁开发了 World Model？
由 tobisamaa（@tobisamaa）开发并维护，当前版本 v2.0.0。
推荐 Skills