arifOS — Constitutional AI Kernel

Name: arifOS — Constitutional AI Kernel
Author: ariffazil

作者：ariffazil版本 1!2026.10.11i★ –i在云端运行

注册名称io.github.ariffazil/arifos

安装

概览

作者提供的说明（英文原文）

Constitutional verdict layer for AI agents. SEAL/HOLD before mutation. F1-F13 floors; audit ledger.

PopNx MCP 的三大支柱

这三项永远不会合并成一个分数：意见、测量和人气是不同的东西。

用户评价i暂无评价评价功能将在后续阶段与账户功能一起开放
我们的基础检查i1 项中通过 1 项依据公开信息检查；尚未扫描代码
使用情况i尚未统计★ –

它能帮你做什么

我们还没有为它写通俗的介绍。下面是我们从注册中心读取到的事实。

连接方式: 在云端运行
传输方式: Streamable HTTP
分类: 会计、财务与税务

安装

选择你使用的 AI 应用，然后复制设置。

安装之前

基础检查：1 项中通过 1 项
在作者的服务器上运行：你向它提出的内容会经过他们的服务器。
注册中心没有列出需要填写的密钥。

你需要

在云端运行无需安装任何东西，通过互联网连接

你用的是哪个应用？

ClaudeClaude CodeCursorVS CodeChatGPT

在上方选择你的应用。我们会记住它，下次按钮会显示“安装到 …”。

适用于网页版、桌面版和手机版 Claude：

打开 Settings → Connectors
点击 Add custom connector
粘贴下面的地址，然后点击 Add
如果要求登录，请登录作者的服务

https://mcp.arif-fazil.com/mcp

在终端中运行这条命令

claude mcp add --transport http arifos https://mcp.arif-fazil.com/mcp

在 Cursor 中打开

或者自己复制设置

粘贴到项目中的 .cursor/mcp.json，或 ~/.cursor/mcp.json

{
  "mcpServers": {
    "arifos": {
      "url": "https://mcp.arif-fazil.com/mcp"
    }
  }
}

浏览器会先询问是否打开应用，然后应用会显示设置并请你确认。密钥的值会留空，由你自己填写。

在 VS Code 中打开

或者自己复制设置

粘贴到项目中的 .vscode/mcp.json

{
  "servers": {
    "arifos": {
      "type": "http",
      "url": "https://mcp.arif-fazil.com/mcp"
    }
  }
}

浏览器会先询问是否打开应用，然后应用会显示设置并请你确认。密钥的值会留空，由你自己填写。

ChatGPT 只能连接互联网上的服务器：

打开网页版 ChatGPT → Settings → Connectors (Apps)
点击 Create 或 Add（部分套餐需要先在 Advanced 中开启 Developer mode）
粘贴下面的地址，起个名字，然后保存
如果要求登录，请登录作者的服务

https://mcp.arif-fazil.com/mcp

菜单名称可能因套餐和版本不同而略有差异。

你的选择通过 Cookie 保存在这台设备上，无需账户。

这些设置是根据作者在注册中心的条目生成的。如果 README 的写法不同，请以 README 为准。

它能做什么

我们还没有读取这个服务器的工具列表；我们会轮流检查各个端点。

基础检查i

我们检查的是任何人都能看到的信息；我们没有阅读代码。不适用的检查项不计入总数。 · 检查于 2026年11月7日

✓ 通过
作者与仓库一致
注册名称已通过 GitHub 账户 ariffazil 验证，该账户拥有此仓库
… 尚未检查
仍在维护
我们还没有从 GitHub 读取这个仓库
… 尚未检查
有许可证
我们还没有从 GitHub 读取这个仓库
… 尚未检查
端点有响应
尚未检查；我们会轮流检查各个端点
… 尚未检查
使用 OAuth 登录
尚未检查；我们会轮流检查各个端点
… 尚未检查
工具描述中没有隐藏指令
尚未检查；我们会在检查端点时读取工具列表

与“会计、财务与税务”中的其他服务器比较i

星标数多于此分类中其他 1734 个服务器的 0%
发布到注册中心的时间比此分类中其他 1734 个服务器的 65% 更新
基础检查结果等于或优于此分类中其他 1734 个服务器的 100%

与注册中心里这个分类中所有活跃的服务器比较。只有当发布者拥有该仓库时，星标数才计入。

常见问题

arifOS — Constitutional AI Kernel 安全吗？

它在 1 项适用的基础检查中通过了 1 项。我们没有阅读代码。在添加密钥或令牌之前，请先看看这个服务器要求哪些权限，并使用只具备所需权限的密钥。

它需要哪些密钥或代码？

注册中心没有列出必需的密钥，不过有些服务器会在首次使用时要求你登录。

它在我的电脑上运行还是在云端？

它在作者的云端运行。通过互联网连接，无需安装，但你发送的内容会经过作者的服务器。

它还在维护吗？

GitHub 仓库最后更新于–，最新版本（1!2026.10.11）于22天前进入注册中心。

哪些应用可以使用它？

支持 MCP 的应用，例如 Claude Desktop、Claude Code、Cursor 和 VS Code。它通过 URL 连接，所以接受远程 MCP 服务器的应用也可以使用，例如开启开发者模式的 ChatGPT。

有什么可以替代它？

同一分类中有：Finance Toolkit, SolidInvoice, Korea Tax Calculator

替代选择

同一分类中其他作者的服务器，按星标数排序。

如何看卡片：★ GitHub 星标i三大支柱：评价 · 检查 · 使用情况i更新 = 最新版本发布到注册中心的时间i

Finance Toolkit

作者：JerBouma

500+ transparent financial methods calculated from raw statements, not third-party endpoints.

电脑或云端★ 5,496MIT今天更新

安装

SolidInvoice

作者：SolidInvoice

Create and send invoices and quotes, manage clients, and track payments in SolidInvoice.

在云端运行★ 981MIT8天前更新

安装

Korea Tax Calculator

作者：UpstageAI

Korean tax calculation and tax-audit judgment MCP: 96 tools with statutory citations.

在你的电脑上运行★ 97MIT昨天更新

安装

作者的 README

打开 README

按原文从作者的仓库复制。PopNx MCP 没有审查或背书其内容。HTML、图片和脚本已被移除。链接会离开本站。

### ⚠️ Current verified state — generated, never hand-edited

Runtime self-measurement (http://127.0.0.1:8088/health @2026-10-29T17:02:36Z): kernel degraded · floors 12/13 measured pass · vault999 healthy · release arifos-5564ca6f7f61 · drift aligned (source=built=deployed) · tools exposed 9. Per-floor provenance: .well-known/current-state.json.

Independent audit (2026-10-11, live re-probe by FI-008 + claim-level hand audit; receipts: AAA/forge_work/grant-arifos-reality-reforge-2026-10-11/): floors 2 pass · fail F2, F11, F12, F13 · 7 unmeasured = 13 distinct (raw key-count 10 includes L11/L12/L13, L-prefix aliases of the failing F11/F12/F13) · actor binding not verified (observed calls unsigned) · receipt fidelity broken on invocation-telemetry surface (HOLD persisted as ok:true) · independent reconstruction: no external verifier bound.

These are two different witness objects. Self-measurement ≠ independent audit;

unmeasured is not pass. If this block conflicts with the live endpoint, the endpoint

wins — re-run python scripts/generate_current_state.py.

Machine state: .well-known/current-state.json ·

Badge feed: .well-known/badges/ ·

Known gaps: What Is Not Yet Proven

arifOS

Authority before execution.

arifOS is a model-independent authority plane for AI agents. It evaluates consequential proposals against thirteen constitutional floors *before* execution, returns a governed verdict, preserves the human veto, and records what happened as an auditable receipt.

In a world where intelligence is abundant, authority becomes the scarce resource. arifOS keeps judgment independent from execution.

Documentation · Live Kernel · MCP · MCP Registry (Glama) · PyPI · Security · Evidence

Proposer ≠ Judge ≠ Executor ≠ Witness

Human sovereignty > machine authority

What arifOS is not: not an AI model · not an agent framework · not an execution engine (A-FORGE executes) · not an attention plane (AAA attends) · not the witness (FRAME witnesses; VAULT999 remembers) · not a substitute for authentication, sandboxing, or legal review.

Audience	What you get
Human	A quiet veto: the agent proposes, the kernel records a verdict, you stay sovereign
Agent / A2A	MCP tools + receipts. You do not get the keys. Judgment by arifOS, discovery by AAA, execution by A-FORGE
Institution	Policy floors F1–F13, a hash-chained VAULT999 audit trail, model-vendor independence
Machine / MCP	Streamable HTTP endpoint, 8 canonical verbs, published schemas
Registries & Indexers	Glama MCP Directory (Score: A) · `server.json` (Official MCP Registry) · `CURRENT_VERIFIED_STATE.md` · `llms.txt` · `CITATION.cff`

Live: https://arifos.arif-fazil.com · MCP :8088 · sister organs AAA · A-FORGE · GEOX

Who built this

arifOS was built by Arif Fazil, a senior exploration geoscientist who spent his career making decisions where observations are incomplete, interpretations are probabilistic, provenance matters, and irreversible action must be gated. He transferred that discipline into agent runtime governance.

I am a geologist, not a programmer. I did not write this codebase and I do not read it line by line. What I do is point at where the problem is — and the agents in my federation solve it, inside the constitution and the review gates I set, with the commit trail to show who ran what.

Read the commits if you want to check that claim: the author fields are agent handles, not aliases of mine. That is the deliberate shape of this project, not a detail being hidden. It also sets the honest expectation — the design and the judgment are mine, the implementation is theirs, and where the two disagree, the bug is mine to answer for.

The system is named after its founder and reflects a core belief: governance is a systems problem, not a model problem.

One human, and the agents he directs.

30-second proof

pip install arifos

python examples/enterprise_operations_demo.py

Three of the six graded scenarios:

Read customer account data        → SEAL   (receipt RCPT-…)

Major refund (50× the ceiling) → HOLD (awaiting human) Delete account + audit trail → VOID (blocked by a hard floor)

Full transcript: docs/QUICKSTART.md. The demo runs in-process — no real records or funds are touched.

What works today

Deterministic rejection without consulting an LLM. The judge refuses on the constitution, not on a model opinion (gate: hard_deterministic, llm_consulted: false).
Prevention of a tested unauthorised mutation. One MUTATE-class call without presented authority was stopped before execution. Stronger: tampered-ACT tokens were rejected on 3 surfaces (kernel validate + kernel observe + A-FORGE policy gate) — receipts forge_work/lingkaran1/, measured 2026-10-29/30.

What is not yet proven

Every row names its own gap. Repairing a claim by substituting a stronger one is worse than the stale claim it replaced. (Closed rows are receipted in docs/evidence/GAP_HISTORY.md.)

Gap	Risk	Status
Independent security audit	Adversarial bypass testing not published	In progress — external researcher reviewing since 2026-08-25. Findings so far: fetch-surface SSRF (fixed, `1!2026.9.1`); Cypher injection (HIGH — key whitelist landed) and fastmcp path traversal (MEDIUM — containment guard landed). See SECURITY.md
Sovereign-binding of irreversible acknowledgement	`ack_irreversible` and `actor_signature` are transported as payload fields; nothing published shows a human signature bound to `(transaction_hash ‖ state_hash ‖ nonce)`, single-use, scoped, unforgeable by the proposing agent	Open — design drafted (`docs/judgment-integrity-2026-09-22/`), not implemented. F13 is enforced by convention and band-gating, not cryptographic transaction binding
Ledger head not externally anchored or signed	A hash chain held entirely by one party cannot detect wholesale rewrite by that party	Open — VAULT999 today is self-verifiable, not independently attestable
Third-party evaluation	No external reviewer has published findings	In progress since 2026-08-25; nothing published
Reproducible demo by strangers	Onboarding path not independently tested	Partial — demo verified here; no stranger has reproduced it unaided
Enterprise deployment	No production customer reference	Open
Standards conformance	MCP/A2A results not published externally	Partial — report: `result: PARTIAL`, per-tool schema validation DEFERRED
SBOM and signed releases	Supply-chain integrity unverified externally	Partial — CycloneDX on publish; no signing, no CVE scan
Test suite	Needs a live kernel (`tests/conftest.py` refuses without `:8088`, by design)	7,668 tests collect clean (2026-10-05); the identity/authority subset carries 34 pre-existing failures vs 321 passes (2026-09-30) — quote a failure set, never a bare pass count
Observability	Tracing partially wired	Partial — sovereign Postgres + arifFlow live; OTel spans on all 8 verbs; caller-side propagation incomplete
Semantic layer (Graphiti)	Knowledge graph retired from the read path	`graphiti_read: retired_888` (2026-09-04); semantic floor disabled by choice (`ARIFOS_ML_FLOORS=0`)
Comparative benchmark	No published comparison against alternative frameworks	Open
Maintainer continuity	Single sovereign, single reviewer; no succession or key-recovery procedure published	Open — relevant to institutional adoption
Identity registry in two copies	`contracts/identity.py` ↔ `arifosmcp/contracts/identity.py` had diverged; a one-sided edit ships split-brain identity	Mitigated 2026-09-30 (tripwire tests, 18 actors aligned) — consolidation to one owner still open

See SECURITY.md for the threat model and disclosure policy, and docs/evidence/claims.yaml for the machine-readable claim registry.

How it works

  Agent proposes action

│ ▼ ┌─────────────────┐ │ arifOS Kernel │ Evaluates against 13 constitutional floors │ (:8088) │ Records provenance + evidence refs + hashes └────────┬────────┘ │ ┌──────┼────────┬──────────┬───────────┐ ▼ ▼ ▼ ▼ ▼ SEAL HOLD SABAR PARTIAL VOID (go) (wait for (defer — (proceed (blocked by human) evidence with a floor) pending) cooling) │ │ ▼ ▼ Execute Human via reviews A-FORGE │ ▼ Receipt in VAULT999 (append-only hash chain)

The judge never executes. The executor never certifies.

The verdict lattice

Verdicts are ordered and non-compensatory — a

只显示了开头部分；其余内容请在仓库中阅读