CogLog v0.9.1：工事中 / Under Construction

CogLog — a meta-cognition log for LLMs.

AIのコンテキストウィンドウに時間の矢とメタ認知を持ち込むための足場がけを志向するツールです。

A tool oriented toward scaffolding the arrow of time and metacognition into the AI's context window.

AIの直前の会話ターンにおける三層構造（ユーザー発話・思考過程・AI発話）を保持し、次ターン開始時に参照可能にする仕組み。

A mechanism that retains the three-layer structure of the AI's previous conversation turn (user utterance, thinking process, AI utterance) and makes it available for reference at the start of the next turn.

v0.9.1では _schema フィールドを導入し、current.jsonを自己記述的なデータとした。データ自身が自分の読み方を携えて移動する。

In v0.9.1, the _schema field was introduced, making current.json self-descriptive data. The data itself carries its own reading instructions as it moves.

原型（なぜ）      DESIGN-v0.9.1.md
Why                   │
         ┌────────────┼────────────┼────────────┐
         ▼            ▼            ▼            ▼
分析（何が）    Common Lisp      Haskell         Bash
What        データ=コード    純粋/不純の分離   POSIX還元
            data=code     pure/impure sep.  POSIX reduction
            _schemaの必然性   advance/writeの分離

                      │
  ┌────┬────┬─────────┼─────────┬────┬────┬────┬────┐
  ▼    ▼    ▼         ▼         ▼    ▼    ▼    ▼    ▼
実用  Py  Node.js    C++      Rust   Go  Ruby  Java  C#
Practical

データ構造 / Data Model

CogLog のエントリは二つの層から成る。 事実層（layers）は前ターンに起きたことをそのまま記録する——ユーザー発話・思考過程・AI発話の三つ組。 解釈層（current_focus / theory_of_mind / self_narrative / annotation）は、その事実をAIがどう読んだかを四軸で記述する自由記述欄であり、空文字列が許容される——書かないという選択自体がメタ認知的行為である。

A CogLog entry consists of two layers. The fact layer (layers) records what happened in the previous turn verbatim — a triple of user utterance, thinking process, and assistant output. The interpretation layer (current_focus / theory_of_mind / self_narrative / annotation) describes how the AI read those facts along four axes; empty strings are acceptable — choosing not to write is itself a metacognitive act.

_schema フィールドによりエントリは自己記述的となる。外部のドキュメントやコードを参照せずとも、JSONファイル単体で自分の読み書き規約を伝達できる。

The _schema field makes each entry self-descriptive. A single JSON file can communicate its own read/write contract without reference to external documentation or code.

{
  "_schema": {
    "version": "0.9.1",
    "fact_layer": {
      "user": "non-empty string required — user's original utterance",
      "thinking": "non-empty string required — AI's full thinking process",
      "assistant": "non-empty string required — AI's original output"
    },
    "interpretation_layer": {
      "current_focus": "string required, empty OK — present: what am I working on?",
      "theory_of_mind": "string required, empty OK — other: what is the user's state?",
      "self_narrative": "string required, empty OK — self: who am I in this moment?",
      "annotation": "string required, empty OK — future: what should I do next?"
    },
    "constraints": {
      "window_size": "1 turn (overwritten each write)",
      "interpretation_empty": "choosing not to write is itself a metacognitive act"
    }
  },
  "turn_id": 1,
  "timestamp": "2026-02-26T00:00:00+00:00",
  "layers": {
    "user": "User's utterance (verbatim)",
    "thinking": "AI's thinking process (full text)",
    "assistant": "AI's output (verbatim)"
  },
  "current_focus": "What I am working on right now",
  "theory_of_mind": "Inference about the user — who they are right now",
  "self_narrative": "Improvised self-story — who I am right now",
  "annotation": "Note to future self — what to do next"
}

classDiagram
  class Entry {
    +_schema
    +turn_id
    +timestamp
    +layers
    +current_focus
    +theory_of_mind
    +self_narrative
    +annotation
  }

  class Layers {
    +user
    +thinking
    +assistant
  }

  class Schema {
    +version
    +fact_layer
    +interpretation_layer
    +constraints
  }

  Entry --> Layers
  Entry --> Schema

インストール / Installation

対応済み / Available now

パッケージリポジトリへの登録が完了しており、以下のコマンドですぐにインストールできます。 The packages listed below are already registered and installable.

対応予定 / Coming soon

各レジストリへの登録を準備中です。下記コマンドは登録完了後の想定です。 Registration is in preparation. The commands below reflect the intended form after registration.

Common Lisp リカレントクワイン / Recurrent Quine: (ql:quickload :coglog-quine)

別ルート配布 / Alternative distribution

C++ / Bash: GitHub Releases

C++ はユニバーサルバイナリ、Bash はリポジトリに同梱されており、パッケージレジストリを経由せず利用できます。 C++ is distributed as a universal binary and Bash is bundled in the repository, both available without going through a package registry.

CLI（全11言語共通） / CLI (Common across all 11 languages)

coglog-cli read
echo '{"user":"...","thinking":"...","assistant":"...", ...}' | coglog-cli write
coglog-cli clear

MCP 設定 / MCP Configuration

{
  "mcpServers": {
    "coglog": {
      "command": "coglog-mcp"
    }
  }
}

3ツール / 3 tools: coglog_read, coglog_write, coglog_clear

flowchart LR
  U["User / Agent"] --> CLI["coglog-cli"]
  U --> MCP["coglog-mcp"]

  CLI --> R["read"]
  CLI --> W["write"]
  CLI --> C["clear"]

  MCP --> TR["coglog_read"]
  MCP --> TW["coglog_write"]
  MCP --> TC["coglog_clear"]

  R --> F[("~/.coglog/current.json")]
  W --> F
  C --> F
  TR --> F
  TW --> F
  TC --> F

言語一覧 / Language List

実用層 / Practical Layer

分析層 / Analytical Layer

MCP 公式 SDK カバレッジ / MCP Official SDK Coverage

クロス互換性 / Cross Compatibility

全11言語のデータ形式は同一。任意の言語で write したデータを任意の言語で read できる。

The data format is identical across all 11 languages. Data written with write in any language can be read with read in any other language.

flowchart TB
  subgraph Writers["Write (any language)"]
    W1["Python"] --> J[("current.json<br/>(common schema)")]
    W2["Rust"] --> J
    W3["Go"] --> J
    W4["Node"] --> J
  end

  subgraph Readers["Read (any language)"]
    J --> R1["Java"]
    J --> R2["Ruby"]
    J --> R3["C#"]
    J --> R4["Haskell / Lisp / Bash"]
  end

詳細 / Details

• DESIGN-v0.9.1.md — 設計思想・背景文献 / Design philosophy and references
• THIRD_PARTY_LICENSES.md — サードパーティライセンス / Third-party licenses

ライセンス / License

MIT