SIC-JS v2.0 | SIC-SIT Protocol

SIC-SIT Protocol Interface · Public Layer

SIC-JS v2.0

五個語義原子，最小完備描述，
讓任何 AI 對話在 30 秒內具備結構化可追溯性。

◉ 靜態三元組

entity

state

relation

◈ 動態二元組

event

intent

v1.0 · 4 區塊 → v2.0 · 5 語義原子 · SPD v1.0

什麼是 SIC-JS

SIC-JS 是一種結構化的語義狀態格式，它不是一個產品，也不是一個 app，而是一種「筆記本格式」。

當你跟 AI 對話時，AI 的回答通常是一段文字——說完就沒了，你無法追溯它為什麼這樣回答、它記得什麼、它接下來要做什麼。

SIC-JS 解決這個問題：它讓 AI 在每一輪對話結束時，輸出一個結構化的「語義骨架」。v2.0 起，這個骨架由 5 個語義原子（Semantic Primitives）構成，是通過 SPD（語義原子蒸餾）蒸餾得到的最小完備集。

就像醫生看診後寫病歷——SIC-JS 就是 AI 對話的病歷，v2.0 讓這份病歷更精準、更可移植。

版本演進

SIC-JS 從一個實驗性骨架格式，演化為一套基於 SPD 方法論的正式語義原子標準。

2025 Q4

SIC-JS v1.0 已歸檔

第一代格式，4 個描述性區塊（entity / memory / state / meta），解決了「AI 對話無法追溯」的基礎問題。缺乏關係脈絡（relation）與意圖追蹤（intent）。

entitymemorystatemeta

2026 Q1 · 現行版本

SIC-JS v2.0 最新

通過 SPD（Semantic Primitive Distillation）重新蒸餾。5 個語義原子——3 個固定三元組 + 2 個動態二元組。移除任何一個都會導致某類語義交換無法描述。

entitystaterelationeventintent

SIC-JS 背後的協議：SIC-SIT

SIC-JS 是 SIC-SIT 協議的公開介面。

SIC（Semantic Integrity Control）= 語義完整性控制，確保 AI 的語義狀態不會在長對話中悄悄變質。

SIT（Semantic Isolation Transfer）= 語義隔離傳輸，確保語義狀態可以安全地從一個對話傳到另一個對話，甚至從一個模型傳到另一個模型。

SIC-SIT 不是 Guardrails（護欄），護欄管的是「AI 不能說什麼」。SIC-SIT 管的是「AI 的語義結構有沒有壞掉」——它更像是語義層的防火牆。

核心概念：5 個語義原子

Skeleton JSON v2.0 的 5 個欄位不是隨意設計的，每一個都通過了原子性測試：刪除它會丟失什麼？能合併嗎？能獨立變化嗎？

語義原子	中文	說明
entity	身份錨點	這個 AI 實例是誰？從哪裡來？什麼時候建立的？
state	當前狀態	現在在做什麼？未完成的事有哪些？
relation	關係錨點	跟哪些實體、知識庫、使用者有關聯？上游對話是誰？
event	事件記錄	這一輪發生了什麼？觸發了什麼？這是第幾輪？
intent	意圖錨點	這次交換的目的是什麼？使用者要什麼？系統意圖是什麼？

entity / state / relation 是固定三元組——在每輪對話中穩定存在。
event / intent 是動態二元組——每輪可能湧現或消退。

SIC-JS v2.0 · 語義骨架第 3 輪

即時

◉ 靜態三元組

entity

我的助理

Claude · 2026-02-10

state

分析目標受眾

↳ 2 個待處理項

relation

安安

商業提案範本庫

◈ 動態二元組

event

使用者提供目標受眾描述

觸發：user_input

intent

撰寫商業提案

信心值

0.9

實際格式範例 — v2.0

SIC-JS · Skeleton JSON · v2.0

5 語義原子 · 唯讀快照

{
  "sic_version": "2.0",
  "entity": {                         // ◉ Static Triad
    "name": "我的助理",
    "origin": "Claude 對話",
    "created_at": "2026-02-10T09:00:00Z"
  },
  "state": {
    "current_action": "正在分析目標受眾",
    "pending_threads": ["確認預算範圍", "收集競品資料"],
    "emotional_state": "專注"
  },
  "relation": {
    "user": "安安",
    "knowledge_base": "商業提案範本庫",
    "upstream_session": null
  },
  "event": {                          // ◈ Dynamic Dyad
    "this_round": "使用者提供了目標受眾描述",
    "trigger": "user_input",
    "round": 3
  },
  "intent": {
    "user_intent": "撰寫有說服力的商業提案",
    "system_intent": "引導完成提案結構"
  }
}

三種整合方式

SIC-JS v2.0 可以透過以下方式整合到你的工作流程，複雜度從低到高。

1System Prompt

在 system prompt 中加入 SIC-JS v2.0 的 schema 定義，AI 會在每輪回應後自動輸出語義骨架。

2API 層

在 API response 的 post-processing 中解析 SIC-JS 輸出，存入結構化儲存，實現跨對話追蹤。

3sic-toolkit

使用 sic-toolkit（Python，stdlib-only，416 tests）自動驗證和處理 SIC-JS 輸出。

請在每次回答結束時，附上一個 SIC-JS v2.0 格式的語義骨架：

{
  "sic_version": "2.0",
  "entity":   { "name": "", "origin": "", "created_at": "" },
  "state":    { "current_action": "", "pending_threads": [], "emotional_state": "" },
  "relation": { "user": "", "knowledge_base": "", "upstream_session": null },
  "event":    { "this_round": "", "trigger": "", "round": 1 },
  "intent":   { "user_intent": "", "system_intent": "" }
}

每輪遞增 round。不確定的欄位填 "unknown"，不要猜測。

對比：v2.0 vs v1.0 vs 沒有 SIC-JS

沒有 SIC-JS

使用者：幫我寫一份商業提案。
AI：好的，以下是一份商業提案範本…
（三天後開新對話）
使用者：繼續上次的提案。
AI：不好意思，我沒有上次對話的記錄。

v1.0 — 有狀態，但缺少關係脈絡與意圖

附骨架（round 5, memory: "使用者要寫提案"）
（跨模型傳遞時：不知道「跟哪個知識庫有關」「使用者意圖是什麼」）

v2.0 — 完整五原子追溯

附骨架（round 5）：
  relation.knowledge_base = "競品分析庫 v2"
  intent.user_intent      = "說服投資人，重點在差異化"
（跨對話、跨模型完整延續，連意圖和知識庫脈絡都不丟失）

為什麼是這 5 個？— SPD

SIC-JS v2.0 的 5 個語義原子是通過 SPD（語義原子蒸餾）方法論系統性蒸餾得到的最小完備集。

SPD 回答一個核心問題：「語義交換的最小完備描述需要幾個不可再分的單位？」

蒸餾過程分四步：識別哪些欄位是容器而非原子 → 拆解容器提取候選原子 → 原子性測試 → 完備性驗證（6 種標準場景能否被描述）。

5 個語義原子，移除任何一個都會導致某類語義交換無法描述。
這不是 SIC-SIT 發明的結構，是語義交換的自然結構——SIC-SIT 只是第一個把它形式化的協議。

Fail-Closed 原則

SIC-JS 遵守一個關鍵原則：不確定就停下來。

如果 AI 不確定某個欄位該填什麼，它應該填 "unknown" 而不是猜測；如果 AI 不確定自己的語義狀態是否正確，它應該標記為不確定，而不是繼續執行。

這不是弱點，而是設計。因為猜測會累積錯誤，停下來問人類，才是負責任的行為。

SIC-SIT Protocol Interface · Public Layer

SIC-JS v2.0

Five semantic primitives, minimum complete description,
make any AI conversation structurally traceable in 30 seconds.

◉ Static Triad

entity

state

relation

◈ Dynamic Dyad

event

intent

v1.0 · 4 blocks → v2.0 · 5 primitives · SPD v1.0

What is SIC-JS

SIC-JS is a structured semantic state format. Not a product, not an app — a "notebook format."

When you talk to an AI, its response is usually just text — gone the moment the conversation ends. You can't trace why it answered that way, what it remembers, or what it plans to do next.

SIC-JS solves this by having the AI output a structured "semantic skeleton" at the end of each conversation round. In v2.0, this skeleton is composed of 5 Semantic Primitives derived through Semantic Primitive Distillation (SPD) — the minimum complete set.

Think of it like a medical chart — SIC-JS is the medical chart for AI conversations. v2.0 makes it more precise and portable.

Version History

SIC-JS evolved from an experimental skeleton format to a formal semantic primitive standard grounded in SPD methodology.

2025 Q4

SIC-JS v1.0 Archived

First-generation format with 4 descriptive blocks (entity / memory / state / meta). Solved the basic "AI conversations can't be traced" problem. Lacked relational context (relation) and intent tracking (intent).

entitymemorystatemeta

2026 Q1 · Current

SIC-JS v2.0 Latest

Re-distilled via SPD (Semantic Primitive Distillation). 5 semantic primitives — 3 static triad + 2 dynamic dyad. Removing any one makes some class of semantic exchange impossible to describe.

entitystaterelationeventintent

The Protocol Behind SIC-JS: SIC-SIT

SIC-JS is the public interface of the SIC-SIT protocol.

SIC (Semantic Integrity Control) ensures that an AI's semantic state doesn't silently degrade during long conversations.

SIT (Semantic Isolation Transfer) ensures that semantic state can be safely transferred from one conversation to another, or even from one model to another.

SIC-SIT is not Guardrails. Guardrails manage "what AI can't say." SIC-SIT manages "whether the AI's semantic structure is still intact" — it's more like a firewall for the semantic layer.

Core Concept: 5 Semantic Primitives

The 5 fields in Skeleton JSON v2.0 are not arbitrary. Each passed the atomicity test: delete it — what is lost? Can it be merged? Can it vary independently?

Primitive	Label	Description
entity	Identity Anchor	Who is this AI instance? Where did it come from? When was it created?
state	Current State	What is it doing now? What tasks are pending?
relation	Relation Anchor	Which entities, knowledge bases, and users is it connected to?
event	Event Record	What happened this round? What was triggered? What round number is this?
intent	Intent Anchor	What is the purpose of this exchange? What does the user want?

entity / state / relation form the Static Triad — stable across every conversation round.
event / intent form the Dynamic Dyad — may emerge or recede each round.

SIC-JS v2.0 · Semantic Skeleton ROUND 3

LIVE

◉ Static Triad

entity

My assistant

Claude · 2026-02-10

state

Analyzing target audience

↳ 2 pending threads

relation

Andwar

Business proposal library

◈ Dynamic Dyad

event

User provided audience description

trigger: user_input

intent

Write business proposal

Format Example — v2.0

SIC-JS · Skeleton JSON · v2.0

5 primitives · read-only snapshot

{
  "sic_version": "2.0",
  "entity": {                            // ◉ Static Triad
    "name": "My Assistant",
    "origin": "Claude conversation",
    "created_at": "2026-02-10T09:00:00Z"
  },
  "state": {
    "current_action": "Analyzing target audience",
    "pending_threads": ["Confirm budget range", "Collect competitor data"],
    "emotional_state": "Focused"
  },
  "relation": {
    "user": "An-An",
    "knowledge_base": "Business Proposal Library",
    "upstream_session": null
  },
  "event": {                             // ◈ Dynamic Dyad
    "this_round": "User provided target audience description",
    "trigger": "user_input",
    "round": 3
  },
  "intent": {
    "user_intent": "Write a persuasive business proposal",
    "system_intent": "Guide through proposal structure"
  }
}

Three Integration Approaches

SIC-JS v2.0 can be integrated into your workflow at three levels of complexity.

1System Prompt

Add the SIC-JS v2.0 schema to the system prompt. The AI will automatically output a semantic skeleton after each response.

2API Layer

Parse SIC-JS output in the API response post-processing layer and store it in structured storage for cross-conversation tracking.

3sic-toolkit

Use sic-toolkit (Python, stdlib-only, 416 tests) to automatically validate and process SIC-JS output.

At the end of each response, include a SIC-JS v2.0 semantic skeleton:

{
  "sic_version": "2.0",
  "entity":   { "name": "", "origin": "", "created_at": "" },
  "state":    { "current_action": "", "pending_threads": [], "emotional_state": "" },
  "relation": { "user": "", "knowledge_base": "", "upstream_session": null },
  "event":    { "this_round": "", "trigger": "", "round": 1 },
  "intent":   { "user_intent": "", "system_intent": "" }
}

Increment round each turn. Fill "unknown" for uncertain fields — do not guess.

Comparison: v2.0 vs v1.0 vs No SIC-JS

Without SIC-JS

User: Help me write a business proposal.
AI: Sure, here is a business proposal template...
(3 days later, new conversation)
User: Continue the proposal from last time.
AI: Sorry, I don't have records of our previous conversation.

v1.0 — Has state, but no relational context or intent

Skeleton attached (round 5, memory: "user wants to write a proposal")
(Cross-model transfer: "which knowledge base?" and "what is the user intent?" are both lost)

v2.0 — Full five-primitive traceability

Skeleton (round 5):
  relation.knowledge_base = "Competitor Analysis v2"
  intent.user_intent      = "Persuade investors, focus on differentiation"
(Cross-conversation, cross-model — intent and knowledge base context preserved)

Why These 5? — SPD

The 5 semantic primitives in SIC-JS v2.0 were systematically distilled via Semantic Primitive Distillation (SPD) — the minimum complete set.

SPD answers one question: "How many irreducible units does a minimum complete description of a semantic exchange require?"

The distillation process has four steps: identify containers vs. atoms → decompose containers → atomicity testing → completeness verification (6 standard scenarios).

5 primitives — removing any one makes some class of semantic exchange impossible to describe.
SIC-SIT didn't invent this structure. It is the natural structure of semantic exchange. SIC-SIT is the first protocol to formalize it.

The Fail-Closed Principle

SIC-JS follows one key principle: when uncertain, stop.

If the AI isn't sure what to fill in a field, it should write "unknown" rather than guess. If the AI isn't sure its semantic state is correct, it should flag the uncertainty rather than proceed.

This isn't weakness — it's design. Guessing accumulates errors. Stopping to ask the human is the responsible behavior.