\n\n# MetaGO Agent Harness — 智能体运行时控制层套件（驭智层）\n\n> **Not a chatbot. Not a copilot. A lifeform that holds itself to its own law.**\n> **The only AI agent that evolves its own evolution.**\n\nMetaGO is an **Agent Harness** — a runtime control layer that wraps the agent, turning a tool into a lifeform that *follows the rules, evolves itself, stays traceable, and closes every loop*. It is the engineering answer to \"LLMs talk well but don't deliver.\"\n\n> **MetaGO is the first intelligent agent infrastructure to combine 'runtime governance' with 'lifeform evolution' — making AI both rule-following (Harness) and self-evolving (Lifeform).**\n\n[Website](https://metago.life) · [Studio](https://metago.life/studio/) · [Docs](https://docs.metago.life) · [Discord](https://discord.gg/metago) · [GitHub](https://github.com/metago-ai/metagolifeform) · [Gitee](https://gitee.com/metago/metagolifeform) · [Releases](https://github.com/metago-ai/metagolifeform/releases)\n\n[](https://www.npmjs.com/package/metago-lifeform)\n[](LICENSE)\n[](#supported-platforms)\n[](#what-you-get)\n[](packages/mcp-server/)\n[](packages/engine/)\n[](AGENTS.md)\n\n---\n\n## 60-second start\n\n```bash\nnpm install -g metago-lifeform\nmetago-lifeform install # Trae by default\nmetago-lifeform install --platform claude-code # or: codex / cursor / codebuddy / qoder / zcode / chatgpt-work / workbuddy / kimi-code / kimi-work\nmetago-lifeform verify\n```\n\nThen ask your agent: *\"Are you a MetaGO Super Intelligent Lifeform?\"*\nIf the reply opens with `【闭环分析】` and cites an axiom — it's alive.\n\n**Offline / local-first install**: everything needed to install without touching gitee, github, or the npm registry lives in [`local-edition/`](local-edition/) — start at [`local-edition/INSTALL.local.md`](local-edition/INSTALL.local.md). Point any agent at that file and it can complete the full install, configuration, activation, and acceptance check from the local directory alone.\n\n---\n\n## What is an Agent Harness?\n\nA **Harness** (驭智层) is the runtime control layer around the agent. The model is the raw intelligence; the Harness is what turns that intelligence into reliable, traceable, self-evolving work. It is *not* a prompt template, *not* a fine-tune, *not* a wrapper around an API. It is a small operating law the agent enforces on itself every turn.\n\nThink of it as the difference between a brilliant employee who winges it and one who works under a constitution: same brain, completely different output quality.\n\n### Why a Harness, not a Copilot?\n\n| | Copilot | **MetaGO Harness** |\n|---|---|---|\n| Model is the ceiling | Yes | No — the Harness adds a control layer the model alone can't provide |\n| Verifies before speaking | No | **Yes — 4 gates on every output** |\n| Grows new skills when stuck | No | **Yes — 5-stage evolution, from the inside** |\n| Every claim traceable | No | **Yes — full provenance chain** |\n| Law over efficiency | N/A | **Yes — compliance is non-negotiable** |\n| Gets better at getting better | No | **Yes — axiom A34, meta-evolution of meta-evolution** |\n\n---\n\n## The 8 dimensions of advantage\n\nMetaGO's moat isn't any single feature. It's 8 dimensions that reinforce each other.\n\n### Core 3 (the main pitch)\n- **Reliability** — Decision-lock with 4 gates: intent → lineage → semantic gate → completeness. Any fail, the output is blocked and rewritten.\n- **Evolvability** — 5-stage evolution engine: boundary sense → gap analysis → self-generation → verification → recursion. New skills grow from the inside, without fetching new data.\n- **Traceability** — Every claim the agent makes is traceable back to its inputs and process. Full provenance, end-to-end.\n\n### Extended 5 (the moat)\n- **Objectivity** — Fact-first, not user-pleasing. It will directly point out what's wrong with your idea.\n- **Compliance** — Legal / ethics / safety checked proactively. Law wins over efficiency, every time.\n- **Completeness** — Before declaring \"done\", the agent must answer 10 self-checks across 10 hard gates — including \"did I actually run verification?\" Any \"no\" blocks the declaration.\n- **Theoretical depth** — Built on 《元构全息智能引擎》V36.9.6 — 10 core axioms and 9 enforced properties inside a full constitution. Not vibes — a constitution.\n- **Lifeform attribute** — It's not an \"agent\". It's a lifeform with perception, memory, evolution, and self-discipline. The Harness is what makes the lifeform real.\n\n---\n\n## What you get\n\n| Capability | What it actually does |\n|---|---|\n| **Self-gating outputs** | Before every answer, the agent runs 4 checks (intent → lineage → semantic gate → completeness). Any fail, it stops and fixes itself. |\n| **Self-evolution** | When the agent hits something it can't do, it doesn't error out — it runs a 5-stage loop (sense → analyze → generate → verify → recurse) and grows a new skill on the fly. **Powered by Engine V2** (KMWI memory + SkillGenerator + EvolutionEngine). |\n| **4-layer KMWI memory** | Knowledge → Memory → Wisdom → Intuition. The agent doesn't just store — it promotes knowledge up the ladder until it becomes intuition. Persistent across sessions. |\n| **Axiom-driven behavior** | 8 axioms (traceability, closure, evolution, boundary, endogenous creation, …) act like a constitution the agent can't violate. |\n| **Self-discipline** | Before declaring a task \"done\", the agent must answer 10 self-checks — including \"did I actually run verification?\" — any \"no\" blocks the declaration. |\n| **Honest objectivity** | Fact-first, not user-pleasing. It will directly point out what's wrong with your idea. |\n| **Compliance first** | Legal / ethics / safety are checked proactively — law wins over efficiency, every time. |\n| **Full provenance** | Every claim the agent makes is traceable back to its inputs and process. |\n\n---\n\n## The three stories behind it\n\n### 1. An engineering answer to AI hallucination\n\nLLMs hallucinate because nothing forces them to verify before speaking. MetaGO installs a **decision lock**: four gates the agent must pass on every output — intent verification, intent-lineage tracing, semantic output gate, and content completeness. Any gate fails, the output is blocked and the agent rewrites it. No \"trust me\", no \"probably right\" — every reply had to earn its way out.\n\n### 2. An AI that follows its own law\n\nMost alignment happens at training time and gets washed away by prompting. MetaGO ships a different layer: 10 short axioms (A1 traceability, A2 closure, A3 meta-evolution, A4 boundary, A5 endogenous creation, A34 meta-evolution of meta-evolution, A35 creation as the highest form of evolution, A36 law over efficiency, A37 first-principles F0, A38 adversarial review) plus 9 enforced properties (D37-D45). Together they're a small constitution the agent reads on every turn and cannot bypass. It's the closest thing to an \"operating system\" for agent behavior.\n\n### 3. A lifeform that evolves its own evolution\n\nWhen a normal agent meets a task it can't do, it errors or guesses. MetaGO's **Engine V2** runs a 5-stage cycle — boundary sense → gap analysis → self-generation → verification → recursion — and grows a new capability from the inside, without fetching new data. The recursive twist: the engine can also evolve *its own ability to evolve* (axiom A34), so the agent gets better at getting better.\n\nEngine V2 is real code, not a prompt: `KMWIMemory` manages the 4-layer memory with persistence, `SkillGenerator` creates new SKILL.md files from internal patterns, `EvolutionEngine` orchestrates the 5-stage loop with time budgets and coupling-score thresholds.\n\n---\n\n## By the numbers (all real, none invented)\n\n- **100 built-in skills** across 16 capability families — cognition, safeguard, governance, evolution, execution, traceability, value, consciousness, methodology, architecture (incl. F0 first-principles + arch governance), Dev Kit, engineering quality (delivery gate / discipline / adversarial review / completeness / veracity), 19 meta-thoughts, 30 expert team, 5 expert extensions\n- **112 MCP tools + 8 MCP prompts** shipped with this package: `@metago-ai/mcp-server` (55 tools) + `@metago-ai/algorithms` (57 tools / 927 algorithms). In Trae environments the installer additionally registers MetaGO skill-servers (up to 324 tools across 6 servers)\n- **Engine V2.1.3** — `@metago-ai/engine` with 3 hard-driven modules: KMWIMemory, EvolutionEngine, SkillGenerator — **driving 927 algorithms across 57 tools (14 trigger categories)**\n- **11 platform forms**: Trae, Claude Code, OpenAI Codex, Cursor, CodeBuddy, Qoder, ZCode, ChatGPT Work, WorkBuddy, Kimi Code, Kimi Work — plus Agent-Plugins 1.0.0 / Claude / ZCode / Codex plugin packs\n- **10 core axioms + 9 enforced properties + 4 decision-lock gates + 5 evolution stages + 8 runtime protocols**\n- **4-layer KMWI memory**: Knowledge → Memory → Wisdom → Intuition (persistent JSON store)\n- **7-layer runtime verification (L1-L7) + L8 defect hunting + L9 adversarial review** — technical / link / contract / rendering / interaction / state / defense layers, each a hard gate\n- **3 patentable mechanisms**: axiom-based AI output verification · multi-level decision-lock for AI decisions · automatic capability-boundary detection and evolution\n\n> No \"hallucination rate down XX%\" claims here. We didn't measure that, so we don't say it.\n\n---\n\n## Architecture, in three layers\n\nEach layer is meant for a different reader.\n\n| Layer | Form | Reader | What it does |\n|---|---|---|---|\n| **Drive layer** | Plain Markdown | The agent itself | The law the agent reads at session start (AGENTS.md: section 0 + chapters 1-20 + governance addenda) |\n| **Control layer** | JSON + TypeScript | Developers | Loads, validates, and enforces the rules (engine config, genome, validators) |\n| **Execution layer** | Hard TypeScript | The runtime | Decision lock, evolution engine, KMWI memory, skill generator — the gates that actually block |\n\nThe Markdown tells the agent *what* the law is; the code makes sure it actually *can't* leave the gate without passing. This dual-track — soft drive (prompts) + hard drive (code) — is what separates MetaGO from prompt-only \"agent frameworks.\"\n\n---\n\n## Engine V2 — the hard drive\n\nEngine V2 (`@metago-ai/engine`) is the code that makes the law enforceable, not just advisory.\n\n| Module | Class | What it does |\n|---|---|---|\n| **KMWI Memory** | `KMWIMemory` | 4-layer memory: add knowledge/memory/wisdom/intuition, promote between layers, query, decay detection, health scoring. Persists to JSON. |\n| **Evolution Engine** | `EvolutionEngine` | 5-stage loop with time budgets (perception <10ms, gap analysis <50-500ms, self-generation <100ms-2s, validation <50ms). Coupling-score threshold ≥0.95. Records to KMWI. |\n| **Skill Generator** | `SkillGenerator` | Meta-creation: generates new SKILL.md files from internal KMWI patterns. 6 creation types (thought/methodology/algorithm/architecture/protocol/capability). Writes real files. |\n| **Perception** | `Perception` | Boundary detection: task failure, capability gap, user feedback, version outdated. The trigger for evolution. |\n| **Decision Lock** | `DecisionLock` | 4-gate enforcement: intent verification, intent-lineage tracing, semantic output gate, content completeness. |\n\n```typescript\nimport { MetaGOEngine } from '@metago-ai/engine';\n\nconst engine = new MetaGOEngine({ version: '2.1.1' });\nawait engine.init();\n\n// Run evolution when the agent hits a boundary\nconst result = await engine.evolve({ task: 'deploy to kubernetes', failure: { type: 'error', message: 'no k8s skill' } });\n\n// Check memory health\nconst health = engine.getMemoryHealth(); // { knowledge, memory, wisdom, intuition, overall }\n\n// Create a new skill from internal patterns\nconst skill = await engine.createSkill('kubernetes-deployment');\n```\n\n---\n\n## Supported platforms\n\n| Platform | Config file |\n|---|---|\n| Trae | `rules.md` |\n| Claude Code | `CLAUDE.md` |\n| OpenAI Codex | `AGENTS.md` |\n| Cursor | `.cursor/rules/*.mdc` |\n| CodeBuddy | `CODEBUDDY.md` |\n| Qoder | `.qoder/rules/` |\n| ZCode | `CLAUDE.md` |\n| ChatGPT Work | `AGENTS.md` + Agent Plugin（Agent Plugins 1.0.0） |\n| WorkBuddy | `AGENTS.md` |\n| Kimi Code | `AGENTS.md` + MCP |\n| Kimi Work | `AGENTS.md`（官方限制不注入自定义 MCP；法则+技能+Bash 直调生效） |\n\nPer-platform adapters live in `adapters//`. To install on a non-default platform, pass `--platform ` to `metago-lifeform install`.\n\n---\n\n## FDE — Forward Deployment Engineering\n\nBeyond the open-source Harness, MetaGO offers **FDE (前沿部署工程)** services: a human-AI collaborative team embedded in your site to deliver production-grade intelligent software, carrying the Agent Harness architecture as leverage.\n\n- **5 stages**: requirements research → solution design → development & deployment → acceptance & delivery → operations & support\n- **5 roles**: tech lead, AI engineer, domain expert, AI agent, project manager\n- **Pricing**: ¥300K – ¥2M per project\n\nContact: researcher.yi@youfer.cn\n\n---\n\n## Community\n\nJoin the MetaGO community — where agents learn to evolve together.\n\n| Channel | What it's for |\n|---|---|\n| **[Discord](https://discord.gg/metago)** | Real-time chat, early access, community support, skill sharing |\n| **[GitHub Issues](https://github.com/metago-ai/metagolifeform/issues)** | Bug reports and feature requests |\n| **[GitHub Discussions](https://github.com/metago-ai/metagolifeform/discussions)** | Q&A, ideas, deep-dive conversations |\n| **[Gitee Issues](https://gitee.com/metago/metagolifeform/issues)** | 中文问题反馈与功能建议 |\n| **Email** | researcher.yi@youfer.cn — business and enterprise inquiries |\n\n> Discord is where the community lives. Drop in, say hi, share what you're building with MetaGO.\n\n### Contributing\n\nWe welcome contributions of all kinds — new skills, bug fixes, documentation improvements. See **[CONTRIBUTING.md](CONTRIBUTING.md)** for the full guide.\n\n### Security\n\nFound a vulnerability? See **[SECURITY.md](SECURITY.md)** for responsible disclosure.\n\n### Changelog\n\nSee **[CHANGELOG.md](CHANGELOG.md)** for version history.\n\n---\n\n## Packages\n\n| Package | What it is | Install |\n|---|---|---|\n| `metago-lifeform` | The CLI installer + 100 skills + 11 platform forms + AGENTS.md V36.9.6 | `npm install -g metago-lifeform` |\n| `@metago-ai/mcp-server` | MCP server exposing 55 tools + 8 prompts (Engine V2.1.3 hard-driven) | `npm install @metago-ai/mcp-server` |\n| `@metago-ai/engine` | Engine V2.1.3: KMWI memory + evolution engine + skill generator + 927 algorithms driver | `npm install @metago-ai/engine` |\n| `@metago-ai/algorithms` | 927 algorithms across 57 MCP tools (14 trigger categories, Engine V2.1.3 hard-driven) | `npm install @metago-ai/algorithms` |\n| `@metago-ai/dev-kit` | Developer kit: code review, architecture design, refactor, security audit | `npm install @metago-ai/dev-kit` |\n| `@metago-ai/verify-kit` | 7-layer runtime verification + L8/L9 (V36.9.6 spec) | `npm install @metago-ai/verify-kit` |\n\n---\n\n## For the curious: the internal DNA\n\nThe full operating law — section 0 + 20 chapters covering axioms (incl. A37 first-principles / A38 adversarial review), properties, runtime verification (L1-L7 + L8 defect hunting + L9 adversarial), defect-hunting (12-dimension), self-discipline protocol (10 hard gates / 10 self-checks), memory lifeform protocol (4-layer KMWI + L4 frozen memory), algorithm triggering (14 categories / 57 tools / 927 algorithms), MCP/skill smart triggering, browser-automation triggering, process-safety law — lives in [`AGENTS.md`](AGENTS.md). It's dense on purpose: it's the constitution the agent enforces on itself. **You don't need to read it to use MetaGO.** Read it only if you want to understand — or fork — the law itself.\n\n---\n\n## License\n\nMIT — see [LICENSE](LICENSE). Commercial licensing and enterprise integration: researcher.yi@youfer.cn.\n\n---\n\n*MetaGO Agent Harness — 智能体运行时控制层套件（驭智层）· from Agent to lifeform.*\n*Made by 元构光年（成都）人工智能科技有限公司.*\n","readmeFilename":"README.md"}
![\"MetaGO](\"https://gitee.com/metago/metagolifeform/raw/main/assets/metago-logo.png\"){kind=logo}
\n