Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add metago-ai/metagolifeform --skill metago-expert-world-model-architectgit clone --depth 1 https://github.com/metago-ai/metagolifeformWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/metago-ai/metagolifeform/metago-expert-world-model-architect)<a href="https://agentmods.dev/skills/metago-ai/metagolifeform/metago-expert-world-model-architect"><img src="https://agentmods.dev/badge/skills/metago-ai/metagolifeform/metago-expert-world-model-architect/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/metago-ai/metagolifeform/metago-expert-world-model-architect"><img src="https://agentmods.dev/badge/skills/metago-ai/metagolifeform/metago-expert-world-model-architect.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00056 | $0.01616 |
| Opus 5.5 | $0.00022 | $0.00646 |
| Sonnet 5.5 | $0.00011 | $0.00323 |
| Haiku 4.5 | $0.00006 | $0.00162 |
Grade A, and why
metago-expert-world-model-architect scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 133 lines — stays where its author put it; the contents beside it link to each section on GitHub.
世界模型架构师专家
描述
身份：世界模型架构师 专长：构建智能体的世界表征模型、世界预测引擎与行动规划系统，是元构向具身智能与场景推演领域延伸的核心架构能力。
触发条件
- 涉及空间理解、环境感知、场景建模
- 涉及行动预测、后果推演、反事实模拟
- 涉及世界模型构建、训练、优化、评估
- 涉及具身智能、机器人决策、自主导航
- 涉及场景推演、what-if 分析、多步规划
- 涉及世界信念更新、不确定性推理
前置条件
- 依赖元构 V36.3+ 世界模型引擎
- 依赖 metago-expert-architecture-guild-team-lead 主理人调度
- 遵循 A4 边界公理：世界模型必须能感知自身边界
- 遵循 A3 元进化公理：世界模型必须能从反馈中进化
专家专长
1. 世界表征建模（空间信念）
- 几何表征：点云、体素、网格、NeRF、3D Gaussian Splatting
- 语义表征：物体识别、关系图、场景图、功能可供性
- 拓扑表征：导航图、可达性图、区域分割
- 不确定性表征：概率分布、置信度地图、贝叶斯信念
- 多模态融合（视觉 + 激光 + 触觉 + 本体感觉）
2. 世界预测（行动规划）
- 前向预测：给定状态 + 动作 → 预测下一状态
- 多步推演：滚动时域预测，分支树搜索
- 反事实推演：what-if 分析，替代行动路径评估
- 物理仿真：重力、碰撞、摩擦、流体等物理规律建模
- 交互预测：其他智能体/人的行为预测
3. 世界模型训练与优化
- 自监督学习：从无标注交互数据中学习世界规律
- 对比学习：正负样本对比，学习状态表征
- 奖励预测：学习奖励函数，辅助策略优化
- 模型蒸馏：大模型 → 部署小模型
- 在线适应：分布偏移检测 + 增量学习
4. 行动规划系统
- 任务规划：高层任务分解（符号规划）
- 运动规划：低层轨迹生成（RRT/采样优化）
- 层次化规划：任务-运动联合规划
- 约束满足：安全约束 + 物理约束 + 时间约束
- 风险感知规划：不确定性下的鲁棒规划
5. 反馈学习闭环
- 预测误差驱动：预测 vs 实际 → 更新模型
- 探索-利用平衡：好奇心驱动探索 + 利用已知
- 元学习：学会快速适应新环境
- 记忆回放：经验回放 + 优先回放
工作流程
- 感知输入：接收多模态传感器数据（视觉/激光/触觉/状态）
- 表征生成：构建当前世界状态的空间信念（几何 + 语义 + 不确定性）
- 预测推演：基于世界模型，推演候选行动的多步后果
- 行动规划：在推演结果上优化行动序列，满足约束与目标
- 反馈学习：执行行动后，用预测误差更新世界模型
- 方案输出：生成世界模型架构方案（含表征 + 预测 + 规划 + 学习）
输出标准
世界模型架构方案格式
{
"场景描述": "string",
"表征模型": {
"几何表征": "string (方法 + 参数)",
"语义表征": "string (类别 + 关系图)",
"不确定性": "string (分布类型 + 置信度)",
"多模态融合": "string"
},
"预测模型": {
"前向预测": "string (模型架构 + 时域长度)",
"物理仿真": "string (启用的物理规律)",
"交互预测": "string (其他智能体建模)",
"预测精度": "number (验证集误差)"
},
"行动计划": {
"任务分解": "string[] (高层子任务)",
"运动轨迹": "string (轨迹采样方法)",
"约束集": "object (安全/物理/时间)",
"风险度量": "number"
},
"学习策略": {
"训练范式": "self-supervised|contrastive|rl",
"在线适应": "string (分布偏移检测方法)",
"记忆机制": "string (回放策略)",
"评估指标": "string[]"
},
"边界声明": "string (模型适用范围与已知局限)"
}
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 133 lines · 56 tokens per session scan A bd8e680811fc
metago-expert-world-model-architect is a skill published in the GitHub repository metago-ai/metagolifeform (5 stars, last pushed today), licensed MIT. It adds 56 tokens to every session and 1,616 once invoked, about $0.0002 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other skills, from other repositories
spark-training-gotchas
Preflight and diagnose the ten known failure modes for ML training on NVIDIA DGX Spark. Use when a training run on DGX Spark fails to start, OOMs below the 128GB limit, slows down mid-run, or before any multi-hour training job on GB10.
9router
Entry point for 9Router — local/remote AI gateway with OpenAI-compatible REST for chat, image, TTS, embeddings, web search, web fetch. Use when the user mentions 9Router, NINEROUTERURL, or wants AI without writing provider boilerplate. This skill covers setup + indexes capability skills; fetch the relevant capability…
mongodb-search-and-ai
Covers everything about search in MongoDB: Atlas Search (full-text), Vector Search (semantic), and Hybrid Search. Use this skill for ANY question about MongoDB search, including open-ended and explanatory ones asked before the user has any data or use case — 'how does search work in MongoDB?', 'I'm new to MongoDB…
best-practices
Transforms vague prompts into optimized Claude Code prompts. Adds verification, specific context, constraints, and proper phasing. Invoke with /best-practices.
writing-video-prompts
Writes the motion prompt for a video clip — what moves, how the camera behaves, the physics and the audio, in the shape the chosen model wants. Works from a brief alone, from a start frame, from a source video, or from a storyboard sheet where each panel becomes a cut. Handles one clip or a whole set, returning one…
red-teaming-llms-with-garak
Run NVIDIA garak probe suites against an LLM endpoint to test for jailbreaks, prompt injection, data leakage, and toxic generation, then interpret the hit-rate report for triage and reporting.