claude-plugin-creo-memories

内容来源：README.md（说明文档） · 原始地址 · 查看安装指南

原始内容

# Creo Memories Plugin

Persistent memory system for Claude Code. Remember context, decisions, and learnings across sessions with semantic search, automatic context delivery, and a 4-scene mental model.

## What's New in v0.34.2

- **Cookbook fix: session-snapshot.md `search` query 必須を明示** — dogfood で発見、 query 空の metadata-only filter は **0 件返却** (`searchType: 'semantic'` / `'hybrid'` どちらでも実装上 query 必須、 v0.34 時点)。 Phase 4 Confirm を `get_memory({id})` 主路線に変更、 Resume / List で query 必須を WARNING 明示
- **SKILL.md frontmatter version drift fix** — `version: 0.34.0` → `0.34.2` (plugin.json `0.34.1` との drift を一気に解消、 0.34.1 を skip)
- **CHANGELOG mis-category fix** — `[0.34.0]` の Changed entry "Spec compliance: license/homepage fields, removed legacy skills.txt" は実際 PR #16 (v0.34.1) で main 着地した内容 → v0.34.1 entry に移動

## What's New in v0.34

- **Internal memory link 構文** ★Creo 特有★ — markdown 本文中で `[label](https://github.com/chronista-club/claude-plugin-creo-memories/tree/HEAD/mem_xxx)` を書くと memory 間 link になる。 graph traversability の基本記法。 ([CREO-156](https://linear.app/chronista/issue/CREO-156))
  - **storage**: shorthand `[label](https://github.com/chronista-club/claude-plugin-creo-memories/tree/HEAD/mem_xxx)` をそのまま保存 (domain hardcode なし、 軽量)
  - **creo-web**: SPA navigation で `/memories/mem_xxx` へ遷移 (page reload なし)
  - **claude.ai / 外部 MCP client**: `get_memory` / `read` / `search` 経由で full URL `https://creo-memories.in/r/mem_xxx` に server-side 自動展開
  - **fenced code block 内**: raw のまま (コード例で `[fake](https://github.com/chronista-club/claude-plugin-creo-memories/tree/HEAD/mem_xxx)` 書ける)
  - **2 layer 分離設計**: creo-ui MarkdownViewer は **不変** (Chronista 全 product の共通基盤、 mem_xxx 知識を組み込まず)、 consumer 側 (creo-memories) で完結
- **Cookbook: link-to-memory** — `cookbooks/link-to-memory.md`、 「派生元 pin / 序破離 chain / dogfood report」 等の use case + 表示 context 別 mechanism 解説
- **SKILL.md セクション 0.5 追加** — Decision Tree 直後に **重要事項** として位置付け (Creo-specific syntax で全 memory 執筆者が知るべき)
- **Phase 1.5+ deferred**: bare `mem_xxx` の Linear-style auto-link / hover preview / autocomplete / backlinks graph
- **Phase 4 deferred**: 他 entity (`atl_xxx` / `tag_xxx` 等) への link は creo-ui に linkResolver prop 追加して plugin 化

## What's New in v0.33

- **Cookbook: Session Snapshot** — `cookbooks/session-snapshot.md`、 session 全体の goal / decisions / open_questions / next-step を **1 memory に pin** する recipe。 cross-worktree / cross-day continuity 用。 既存 `record_work_log` (event 単位) と粒度分離、 `onboarding.md` の resume と pair (== Capture/Structure/Pin/Confirm/Resume の 5 motion を既存 70 tool でレシピ化、 新 tool 追加なし)
- **Onboarding cookbook 増補** — 「前 session の session-snapshot から resume」 chapter を **step 0** として追加 (`search(tags:['session-snapshot'])` → next_step pickup → open_questions 提示)
- **Stop hook 強化** — session 終了前 checklist に session-snapshot pin nudge を 1 行追加。 「session goal + decisions が連続したら 1 memory に pin」 を passive nudge
- **No skill rename** — `creo-memories` skill 名は維持、 既存メンタルモデル (4-scene / 序破離 / decision tree) と整合性最優先で機能取り込み

## What's New in v0.30

- **API Redesign RFC v2** — Open Questions 4 件 answer + Breaking Change Inventory:
  - **Q1** `edge` first-class resource 採用 (link verb 抽象は却下)
  - **Q2** `transaction({ ops })` 専用 verb 採用 (write 配列 overload は却下)、 `$N` placeholder で前 op 結果参照
  - **Q3** subscribe channel = pull default + webhook + SSE 段階提供
  - **Q4** Pre-save Detection は `write` option 維持 (専用 hook 化却下)
- **Breaking Change Inventory** — v1.0.0 で削除される ~50 legacy tool の完全 list (60+ entry table)
- **Implementation Sprint Plan** — 9-week roadmap (Sprint 1-8)、 server-side migration の concrete plan
- **Q5 (fetch-by-ID) は v0.30 で完了** — creo-memories PR #353 で `get_memory` server-side 実装、 plugin RFC priority bump 達成

## What's New in v0.29

- **Cookbook: Memory ID で fetch** — `cookbooks/fetch-memory-by-id.md`、 dogfood で発覚した API gap (id 直 fetch tool 無し) の workaround pattern 4 種を documented:
  - search verbose mode (推奨、 認証込みで private も)
  - public memory の HTTP `/api/public/r/<id>`
  - get_provenance preview (1 行のみ)
  - concept_get_by_memory (metadata のみ)
- **RFC v1.5 priority bump** — `read({resource:'memory', id})` を **server-side 着手最優先** に位置付け、 v0.29 priority section 追加
- **SKILL.md tool inventory 拡充** (imp-003 完了) — 4-scene 別に **全 70 tool** を表形式で明示、 言及率 30% → ~80% に向上 (navigability 改善)

## What's New in v0.28

- **`scripts/quarterly-loop.sh`** — 90-day cadence の data prep + meta-loop checklist。 plugin commit / release history + Layer 1 / cross-cycle convergence trend prep + counterfactual + strategic direction の 10 section scaffold (~3h reasoning は agent)
- **`creo-memories/reference/api-redesign-rfc.md`** — 70 → 11 tool redesign の **formal spec phase**。 v0.23 proposal を superceding:
  - 11 tool の **JSON schema** (input / output / discriminated union)
  - **Migration matrix** (50+ legacy tool → new mapping table)
  - **3 phase migration** (v0.24 並立 → v0.25 deprecation → v1.0.0 removal)
  - **Test scenarios** (各 tool 5+ 件)
  - **Open questions** for RFC v2 (edge first-class? batch transaction? channel impl?)

## What's New in v0.27

- **jaq drop-in 自動検出** — 全 hook + script に `JQ=$(command -v jaq 2>/dev/null || command -v jq)` を inline 化。 jaq install 済なら自動で 5-10x 速い path、 未 install なら jq fallback
- **Atlas auto-inference** — `scripts/infer-atlas.sh` で current branch (Linear-style: `mako/creo-103-...` → `creo-memories`) + cwd basename + package.json name から atlas 推定。 SessionStart hook が自動的に「💡 Suggested Atlas: ...」を提示、 mcp__creo-memories__remember 等の atlasId を埋めるヒント
- **Weekly + Biweekly loop wrapper** — `scripts/{weekly,biweekly}-loop.sh` で metric prep + checklist scaffold 自動化。 daily-loop.sh と合わせて 3 cadence の auto-prep 完備 (quarterly のみ依然 manual)

## What's New in v0.26

- **Plugin tool invocation instrumentation** — `PostToolUse` hook が `mcp__*creo-memories__*` 呼び出しを log:
  - 出力先: `~/.claude/creo-memories-invocation.log`
  - dogfood で観測: 「自分が plugin tool 何回呼んだか」 を biweekly loop で trend 化可能
- **Daily loop wrapper script** — `scripts/daily-loop.sh` で Layer 1 memory health + invocation stats を 1 command で集約
- **Invocation stats utility** — `scripts/invocation-stats.sh today|week|month|--since` で期間 filter + tool ranking (Markdown 表)
- **jaq drop-in note** — `jq` の Rust 製代替 [`jaq`](https://github.com/01mf02/jaq) は hook 高頻度実行で起動 5-10x 速い、 syntax 互換、 `brew install jaq` で導入可

## What's New in v0.25

- **Self-Improvement Loop** — ecosystem (plugin / 本体 / docs / external / skill 自身) を周期的 audit する仕組み
  - 4 cadence: daily (5 min) / weekly (20 min) / **biweekly (75 min, primary)** / quarterly (3 h)
  - 設計原則 (8 axiom): hierarchical / Loop Dashboard pattern / ICE prioritization / incident-triggered ad-hoc / meta-loop / convergence / counterfactual / action chain closure
  - Slash command `/creo-memories:improvement-loop [cadence]` (manual)
  - SessionStart hook reminder (passive nudge)
  - Cron-schedulable (autonomous via `/loop` skill)
  - 詳細: [creo-memories/reference/improvement-loop/](https://github.com/chronista-club/claude-plugin-creo-memories/blob/HEAD/creo-memories/reference/improvement-loop/README.md)

## What's New in v0.24

- **Active hooks** — passive echo hook → 真に動作する nudge hook へ:
  - `PreToolUse(Write)` で `*/memory/*.md` 検出時に Layer 判定 prompt
  - `UserPromptSubmit` で decision keyword (決定 / confirmed / done / merged 等) 検出 → remember 提案
  - `Stop` で session 終了前 checklist (remember / record_work_log / ttl 昇格 / complete_with_context)
- **Memory templates** — 7 scaffold (Layer 1: feedback / reference-card / project-canon、 Layer 2: decision-record / bug-fix / phase-completion / work-log)

## What's New in v0.23

- **2-layer architecture** — Local file canon (Layer 1) + Cloud trace-archive (Layer 2) の役割分担を明示
- **4-scene mental model** — `/memories /atlas /views /actions` の 4 scene で operation を整理
- **Decision tree** at skill top — どこに何を書くかが 1 query で決まる
- **Trigger patterns + Anti-patterns** で活用駆動
- **5 cookbooks** (Phase 完了 / Bug fix / Decision Record / Cycle close / Onboarding)
- **API Redesign Proposal** — 70 tool → 11 tool 将来構想 doc 化

## 2-Layer Architecture (重要)

Creo Memories は **2 layer 並立**:

| Layer | 場所 | 役割 |
|---|---|---|
| **Layer 1 — Local Canon** | `~/.claude/projects/<project>/memory/*.md` (markdown file) | **不変方針 / cross-project rule / reference card**。 緩慢に変化、 solo-authored、 MEMORY.md index で管理 |
| **Layer 2 — Cloud Trace** | `mcp.creo-memories.in` (`mcp__creo-memories__*` tools) | **動的 project state / 出来事 trace / multi-agent collaboration**。 速く変化、 共有、 semantic search、 Atlas / Concept で structure |

### どちらに書くか?

```
Q1: 「自分 / プロジェクトが常に従う方針」か?
  Yes → Layer 1 (local file)
  No → Q2

Q2: 「ある時点で起こった事実 / 決定 / 現状」か?
  Yes → Layer 2 (cloud)
  No → 書かなくて OK

Q3 (補助): multi-agent / multi-session で参照?
  Yes → Layer 2 (cloud)
  Solo → Layer 1
```

**迷ったら Layer 2**。 Layer 1 は厳しめに gate (MEMORY.md は session start で auto-load → noise 厳禁)。

詳細: [creo-memories/reference/decision-tree.md](https://github.com/chronista-club/claude-plugin-creo-memories/blob/HEAD/creo-memories/reference/decision-tree.md)

## 4-Scene Mental Model

Layer 2 (cloud) の operation は 4 scene で組織化:

```
┌─────────────────┬──────────────────┐
│ /memories       │ /atlas           │
│ data layer      │ structure layer  │
│ (memory CRUD)   │ (整理 / 共有)    │
├─────────────────┼──────────────────┤
│ /views          │ /actions         │
│ perspective     │ motion layer     │
│ (compass/story) │ (todo/link/log)  │
└─────────────────┴──────────────────┘
```

各 scene の playbook:
- [/memories](https://github.com/chronista-club/claude-plugin-creo-memories/blob/HEAD/creo-memories/reference/scenes/memories.md) — data
- [/atlas](https://github.com/chronista-club/claude-plugin-creo-memories/blob/HEAD/creo-memories/reference/scenes/atlas.md) — structure
- [/views](https://github.com/chronista-club/claude-plugin-creo-memories/blob/HEAD/creo-memories/reference/scenes/views.md) — perspective
- [/actions](https://github.com/chronista-club/claude-plugin-creo-memories/blob/HEAD/creo-memories/reference/scenes/actions.md) — motion

## Quick Start Cookbook

### A. 設計決定を保存

```
remember({
  content: '# {決定内容}\n## 理由\n...',
  category: 'decision',
  status: 'done',
  conceptIds: ['{tag}'],
  atlasId: '{project}'
})
```

→ Pre-save Detection で類似 memory 提案。 supersede 必要なら同 call の `supersedes:[mem_old]`。

### B. PR / Linear と pair (mandate)

```
remember({ content, category:'task', status:'in-progress', atlasId })
+ link_external({ memoryId, externalSystem:'linear', externalId:'CREO-XXX' })
+ link_external({ memoryId, externalSystem:'github', externalId:'PR-XXX' })
完了時: complete_with_context({ memoryId, resultSummary, externalUrl })
```

### C. agent 間 comm を残す

```
record_work_log({
  type: 'decision' | 'message' | 'question' | 'progress' | 'review',
  sender, receiver, content, projectId
})
```

### D. Cycle close

```
get_profile() / project_progress(atlasId)
generate_compass(atlasId) → snapshot を remember
detect_processes() → create_process で chain narrative 化
memory_health() → stale 整理
```

詳細: [Cookbook 一覧](https://github.com/chronista-club/claude-plugin-creo-memories/blob/HEAD/creo-memories/SKILL.md#5-cookbook-具体的-recipe)

## Features

- **Context Engine v3.0** — session start で過去 memory + 未完 todo 自動提供
- **Semantic Search** — 意味検索 (Qdrant + embedding)
- **Atlas + Concept** — 階層 tree + 統合分類 (categories/labels/tags)
- **Process / Compass / Story** — chain narrative + LLM 自動 summary
- **Annotation** — thread 型 review (comment/question/concern/suggestion/approval)
- **Work Log** — agent 間 comm の persist
- **Subscription** — push 型変更通知
- **Team / Shared Context** — multi-user 共有
- **2-Layer + 4-Scene** — mental model で活用駆動 (v0.23 NEW)

## Installation

### From GitHub

```bash
/install chronista-club/claude-plugin-creo-memories
```

### Manual Setup

```bash
claude mcp add --transport http creo-memories https://mcp.creo-memories.in
```

Or add to `.mcp.json`:

```json
{
  "mcpServers": {
    "creo-memories": {
      "type": "http",
      "url": "https://mcp.creo-memories.in"
    }
  }
}
```

## MCP Tools (現状: ~70 tools)

各 scene 別 tool 一覧:

### `/memories` scene (data)
| Tool | Description |
|------|-------------|
| `remember` | Memory 保存 (Pre-save Detection 付き、 supersedes / extends / derives 指定可) |
| `search` | Semantic + structured search (scope / atlas / concept filter) |
| `update_memory` | 部分更新 (楽観的 lock 対応) |
| `forget` | 削除 |
| `annotate` / `get_annotations` / `reply_annotation` | Thread 型注釈 |
| `get_provenance` / `get_relations` | 関係 graph (Mermaid) |

### `/atlas` scene (structure)
| Tool | Description |
|------|-------------|
| `create_atlas` / `list_atlas` / `get_atlas_tree` / `update_atlas` / `delete_atlas` | Atlas 操作 |
| `invite_to_atlas` / `share_atlas` / `unshare_atlas` / `list_shared_atlas` | 招待 / 共有 |
| `concept_*` (8 tools) | Concept 統合分類 (categories/labels/tags) |
| `team_*` (4 tools) | Team 管理 |

### `/views` scene (perspective)
| Tool | Description |
|------|-------------|
| `generate_compass` | Atlas 全体の Concept 別 summary |
| `generate_story` | Atlas narrative 生成 |
| `create_process` / `get_process` / `detect_processes` | Memory chain を Process 化 |
| `memory_health` | Stale / broken-link / 偏り検出 |
| `get_profile` | 自分の活動 profile |
| `project_progress` | 進捗 report |
| `system_health` / `diagnose` / `search_logs` | 健全性 / log |

### `/actions` scene (motion)
| Tool | Description |
|------|-------------|
| `create_todo` / `list_todos` / `update_todo` / `complete_todo` / `delete_todo` | Todo |
| `link_external` / `complete_with_context` / `find_by_external` | External (Linear/GitHub) |
| `subscribe_memories` / `unsubscribe_memories` / `list_subscriptions` / `check_notifications` | Push 通知 |
| `record_work_log` / `search_work_logs` | Agent comm trace |
| `update_presence` / `get_presence` | Presence |
| `create_shared_context` / `list_shared_contexts` / `get_shared_context` / `add_to_shared_context` / `join_shared_context` / `leave_shared_context` | Shared context |
| `end_session` | Session 終了 |

詳細: [creo-memories/reference/mcp-tools.md](https://github.com/chronista-club/claude-plugin-creo-memories/blob/HEAD/creo-memories/reference/mcp-tools.md)

### Future API (v0.24+ 予定)

70 tool → 6 core verbs + 5 named conveniences = **11 tools** へリデザイン提案中:

```
Core: read / write / remove / query / transform / subscribe
Named: remember / recall / annotate / complete_with_context / record_work_log / end_session
```

詳細: [creo-memories/reference/api-redesign.md](https://github.com/chronista-club/claude-plugin-creo-memories/blob/HEAD/creo-memories/reference/api-redesign.md)

## Setup / Authentication

1. Plugin install (`/install chronista-club/claude-plugin-creo-memories`)
2. 初回利用時 OAuth (Auth0) prompt — Google / GitHub login
3. 以後 auto

API Key (programmatic):
- `generate_api_key` で発行
- `Authorization: Bearer <key>` header で使用

## Requirements

- Claude Code
- Creo Memories account (free tier)

## Links

| | URL |
|---|-----|
| MCP Endpoint | `https://mcp.creo-memories.in` |
| Web Viewer | `https://creo-memories.in` |
| GitHub | `https://github.com/chronista-club/claude-plugin-creo-memories` |

## License

MIT

Creo Memories Plugin

Persistent memory system for Claude Code. Remember context, decisions, and learnings across sessions with semantic search, automatic context delivery, and a 4-scene mental model.

What's New in v0.34.2

Cookbook fix: session-snapshot.md search query 必須を明示 — dogfood で発見、 query 空の metadata-only filter は 0 件返却 (searchType: 'semantic' / 'hybrid' どちらでも実装上 query 必須、 v0.34 時点)。 Phase 4 Confirm を get_memory({id}) 主路線に変更、 Resume / List で query 必須を WARNING 明示
SKILL.md frontmatter version drift fix — version: 0.34.0 → 0.34.2 (plugin.json 0.34.1 との drift を一気に解消、 0.34.1 を skip)
CHANGELOG mis-category fix — [0.34.0] の Changed entry "Spec compliance: license/homepage fields, removed legacy skills.txt" は実際 PR #16 (v0.34.1) で main 着地した内容 → v0.34.1 entry に移動

What's New in v0.34

Internal memory link 構文 ★Creo 特有★ — markdown 本文中で [label](https://github.com/chronista-club/claude-plugin-creo-memories/tree/HEAD/mem_xxx) を書くと memory 間 link になる。 graph traversability の基本記法。 (CREO-156)
- storage: shorthand [label](https://github.com/chronista-club/claude-plugin-creo-memories/tree/HEAD/mem_xxx) をそのまま保存 (domain hardcode なし、軽量)
- creo-web: SPA navigation で /memories/mem_xxx へ遷移 (page reload なし)
- claude.ai / 外部 MCP client: get_memory / read / search 経由で full URL https://creo-memories.in/r/mem_xxx に server-side 自動展開
- fenced code block 内: raw のまま (コード例で [fake](https://github.com/chronista-club/claude-plugin-creo-memories/tree/HEAD/mem_xxx) 書ける)
- 2 layer 分離設計: creo-ui MarkdownViewer は不変 (Chronista 全 product の共通基盤、 mem_xxx 知識を組み込まず)、 consumer 側 (creo-memories) で完結
Cookbook: link-to-memory — cookbooks/link-to-memory.md、「派生元 pin / 序破離 chain / dogfood report」等の use case + 表示 context 別 mechanism 解説
SKILL.md セクション 0.5 追加 — Decision Tree 直後に 重要事項 として位置付け (Creo-specific syntax で全 memory 執筆者が知るべき)
Phase 1.5+ deferred: bare mem_xxx の Linear-style auto-link / hover preview / autocomplete / backlinks graph
Phase 4 deferred: 他 entity (atl_xxx / tag_xxx 等) への link は creo-ui に linkResolver prop 追加して plugin 化

What's New in v0.33

Cookbook: Session Snapshot — cookbooks/session-snapshot.md、 session 全体の goal / decisions / open_questions / next-step を 1 memory に pin する recipe。 cross-worktree / cross-day continuity 用。既存 record_work_log (event 単位) と粒度分離、 onboarding.md の resume と pair (== Capture/Structure/Pin/Confirm/Resume の 5 motion を既存 70 tool でレシピ化、新 tool 追加なし)
Onboarding cookbook 増補 — 「前 session の session-snapshot から resume」 chapter を step 0 として追加 (search(tags:['session-snapshot']) → next_step pickup → open_questions 提示)
Stop hook 強化 — session 終了前 checklist に session-snapshot pin nudge を 1 行追加。「session goal + decisions が連続したら 1 memory に pin」を passive nudge
No skill rename — creo-memories skill 名は維持、既存メンタルモデル (4-scene / 序破離 / decision tree) と整合性最優先で機能取り込み

What's New in v0.30

API Redesign RFC v2 — Open Questions 4 件 answer + Breaking Change Inventory:
- Q1 edge first-class resource 採用 (link verb 抽象は却下)
- Q2 transaction({ ops }) 専用 verb 採用 (write 配列 overload は却下)、 $N placeholder で前 op 結果参照
- Q3 subscribe channel = pull default + webhook + SSE 段階提供
- Q4 Pre-save Detection は write option 維持 (専用 hook 化却下)
Breaking Change Inventory — v1.0.0 で削除される ~50 legacy tool の完全 list (60+ entry table)
Implementation Sprint Plan — 9-week roadmap (Sprint 1-8)、 server-side migration の concrete plan
Q5 (fetch-by-ID) は v0.30 で完了 — creo-memories PR #353 で get_memory server-side 実装、 plugin RFC priority bump 達成

What's New in v0.29

Cookbook: Memory ID で fetch — cookbooks/fetch-memory-by-id.md、 dogfood で発覚した API gap (id 直 fetch tool 無し) の workaround pattern 4 種を documented:
- search verbose mode (推奨、認証込みで private も)
- public memory の HTTP /api/public/r/<id>
- get_provenance preview (1 行のみ)
- concept_get_by_memory (metadata のみ)
RFC v1.5 priority bump — read({resource:'memory', id}) を server-side 着手最優先 に位置付け、 v0.29 priority section 追加
SKILL.md tool inventory 拡充 (imp-003 完了) — 4-scene 別に 全 70 tool を表形式で明示、言及率 30% → ~80% に向上 (navigability 改善)

What's New in v0.28

scripts/quarterly-loop.sh — 90-day cadence の data prep + meta-loop checklist。 plugin commit / release history + Layer 1 / cross-cycle convergence trend prep + counterfactual + strategic direction の 10 section scaffold (~3h reasoning は agent)
creo-memories/reference/api-redesign-rfc.md — 70 → 11 tool redesign の formal spec phase。 v0.23 proposal を superceding:
- 11 tool の JSON schema (input / output / discriminated union)
- Migration matrix (50+ legacy tool → new mapping table)
- 3 phase migration (v0.24 並立 → v0.25 deprecation → v1.0.0 removal)
- Test scenarios (各 tool 5+ 件)
- Open questions for RFC v2 (edge first-class? batch transaction? channel impl?)

What's New in v0.27

jaq drop-in 自動検出 — 全 hook + script に JQ=$(command -v jaq 2>/dev/null || command -v jq) を inline 化。 jaq install 済なら自動で 5-10x 速い path、未 install なら jq fallback
Atlas auto-inference — scripts/infer-atlas.sh で current branch (Linear-style: mako/creo-103-... → creo-memories) + cwd basename + package.json name から atlas 推定。 SessionStart hook が自動的に「💡 Suggested Atlas: ...」を提示、 mcp__creo-memories__remember 等の atlasId を埋めるヒント
Weekly + Biweekly loop wrapper — scripts/{weekly,biweekly}-loop.sh で metric prep + checklist scaffold 自動化。 daily-loop.sh と合わせて 3 cadence の auto-prep 完備 (quarterly のみ依然 manual)

What's New in v0.26

Plugin tool invocation instrumentation — PostToolUse hook が mcp__*creo-memories__* 呼び出しを log:
- 出力先: ~/.claude/creo-memories-invocation.log
- dogfood で観測: 「自分が plugin tool 何回呼んだか」を biweekly loop で trend 化可能
Daily loop wrapper script — scripts/daily-loop.sh で Layer 1 memory health + invocation stats を 1 command で集約
Invocation stats utility — scripts/invocation-stats.sh today|week|month|--since で期間 filter + tool ranking (Markdown 表)
jaq drop-in note — jq の Rust 製代替 jaq は hook 高頻度実行で起動 5-10x 速い、 syntax 互換、 brew install jaq で導入可

What's New in v0.25

Self-Improvement Loop — ecosystem (plugin / 本体 / docs / external / skill 自身) を周期的 audit する仕組み
- 4 cadence: daily (5 min) / weekly (20 min) / biweekly (75 min, primary) / quarterly (3 h)
- 設計原則 (8 axiom): hierarchical / Loop Dashboard pattern / ICE prioritization / incident-triggered ad-hoc / meta-loop / convergence / counterfactual / action chain closure
- Slash command /creo-memories:improvement-loop [cadence] (manual)
- SessionStart hook reminder (passive nudge)
- Cron-schedulable (autonomous via /loop skill)
- 詳細: creo-memories/reference/improvement-loop/

What's New in v0.24

Active hooks — passive echo hook → 真に動作する nudge hook へ:
- PreToolUse(Write) で */memory/*.md 検出時に Layer 判定 prompt
- UserPromptSubmit で decision keyword (決定 / confirmed / done / merged 等) 検出 → remember 提案
- Stop で session 終了前 checklist (remember / record_work_log / ttl 昇格 / complete_with_context)
Memory templates — 7 scaffold (Layer 1: feedback / reference-card / project-canon、 Layer 2: decision-record / bug-fix / phase-completion / work-log)

What's New in v0.23

2-layer architecture — Local file canon (Layer 1) + Cloud trace-archive (Layer 2) の役割分担を明示
4-scene mental model — /memories /atlas /views /actions の 4 scene で operation を整理
Decision tree at skill top — どこに何を書くかが 1 query で決まる
Trigger patterns + Anti-patterns で活用駆動
5 cookbooks (Phase 完了 / Bug fix / Decision Record / Cycle close / Onboarding)
API Redesign Proposal — 70 tool → 11 tool 将来構想 doc 化

2-Layer Architecture (重要)

Creo Memories は 2 layer 並立:

Layer	場所	役割
Layer 1 — Local Canon	`~/.claude/projects/<project>/memory/*.md` (markdown file)	不変方針 / cross-project rule / reference card。緩慢に変化、 solo-authored、 MEMORY.md index で管理
Layer 2 — Cloud Trace	`mcp.creo-memories.in` (`mcp__creo-memories__*` tools)	動的 project state / 出来事 trace / multi-agent collaboration。速く変化、共有、 semantic search、 Atlas / Concept で structure

どちらに書くか?

Q1: 「自分 / プロジェクトが常に従う方針」か?
  Yes → Layer 1 (local file)
  No → Q2

Q2: 「ある時点で起こった事実 / 決定 / 現状」か?
  Yes → Layer 2 (cloud)
  No → 書かなくて OK

Q3 (補助): multi-agent / multi-session で参照?
  Yes → Layer 2 (cloud)
  Solo → Layer 1

迷ったら Layer 2。 Layer 1 は厳しめに gate (MEMORY.md は session start で auto-load → noise 厳禁)。

詳細: creo-memories/reference/decision-tree.md

4-Scene Mental Model

Layer 2 (cloud) の operation は 4 scene で組織化:

┌─────────────────┬──────────────────┐
│ /memories       │ /atlas           │
│ data layer      │ structure layer  │
│ (memory CRUD)   │ (整理 / 共有)    │
├─────────────────┼──────────────────┤
│ /views          │ /actions         │
│ perspective     │ motion layer     │
│ (compass/story) │ (todo/link/log)  │
└─────────────────┴──────────────────┘

各 scene の playbook:

/memories — data
/atlas — structure
/views — perspective
/actions — motion

Quick Start Cookbook

A. 設計決定を保存

remember({
  content: '# {決定内容}\n## 理由\n...',
  category: 'decision',
  status: 'done',
  conceptIds: ['{tag}'],
  atlasId: '{project}'
})

→ Pre-save Detection で類似 memory 提案。 supersede 必要なら同 call の supersedes:[mem_old]。

B. PR / Linear と pair (mandate)

remember({ content, category:'task', status:'in-progress', atlasId })
+ link_external({ memoryId, externalSystem:'linear', externalId:'CREO-XXX' })
+ link_external({ memoryId, externalSystem:'github', externalId:'PR-XXX' })
完了時: complete_with_context({ memoryId, resultSummary, externalUrl })

C. agent 間 comm を残す

record_work_log({
  type: 'decision' | 'message' | 'question' | 'progress' | 'review',
  sender, receiver, content, projectId
})

D. Cycle close

get_profile() / project_progress(atlasId)
generate_compass(atlasId) → snapshot を remember
detect_processes() → create_process で chain narrative 化
memory_health() → stale 整理

詳細: Cookbook 一覧

Features

Context Engine v3.0 — session start で過去 memory + 未完 todo 自動提供
Semantic Search — 意味検索 (Qdrant + embedding)
Atlas + Concept — 階層 tree + 統合分類 (categories/labels/tags)
Process / Compass / Story — chain narrative + LLM 自動 summary
Annotation — thread 型 review (comment/question/concern/suggestion/approval)
Work Log — agent 間 comm の persist
Subscription — push 型変更通知
Team / Shared Context — multi-user 共有
2-Layer + 4-Scene — mental model で活用駆動 (v0.23 NEW)

Installation

From GitHub

/install chronista-club/claude-plugin-creo-memories

Manual Setup

claude mcp add --transport http creo-memories https://mcp.creo-memories.in

Or add to .mcp.json:

{
  "mcpServers": {
    "creo-memories": {
      "type": "http",
      "url": "https://mcp.creo-memories.in"
    }
  }
}

MCP Tools (現状: ~70 tools)

各 scene 別 tool 一覧:

`/memories` scene (data)

Tool	Description
`remember`	Memory 保存 (Pre-save Detection 付き、 supersedes / extends / derives 指定可)
`search`	Semantic + structured search (scope / atlas / concept filter)
`update_memory`	部分更新 (楽観的 lock 対応)
`forget`	削除
`annotate` / `get_annotations` / `reply_annotation`	Thread 型注釈
`get_provenance` / `get_relations`	関係 graph (Mermaid)

`/atlas` scene (structure)

Tool	Description
`create_atlas` / `list_atlas` / `get_atlas_tree` / `update_atlas` / `delete_atlas`	Atlas 操作
`invite_to_atlas` / `share_atlas` / `unshare_atlas` / `list_shared_atlas`	招待 / 共有
`concept_*` (8 tools)	Concept 統合分類 (categories/labels/tags)
`team_*` (4 tools)	Team 管理

`/views` scene (perspective)

Tool	Description
`generate_compass`	Atlas 全体の Concept 別 summary
`generate_story`	Atlas narrative 生成
`create_process` / `get_process` / `detect_processes`	Memory chain を Process 化
`memory_health`	Stale / broken-link / 偏り検出
`get_profile`	自分の活動 profile
`project_progress`	進捗 report
`system_health` / `diagnose` / `search_logs`	健全性 / log

`/actions` scene (motion)

Tool	Description
`create_todo` / `list_todos` / `update_todo` / `complete_todo` / `delete_todo`	Todo
`link_external` / `complete_with_context` / `find_by_external`	External (Linear/GitHub)
`subscribe_memories` / `unsubscribe_memories` / `list_subscriptions` / `check_notifications`	Push 通知
`record_work_log` / `search_work_logs`	Agent comm trace
`update_presence` / `get_presence`	Presence
`create_shared_context` / `list_shared_contexts` / `get_shared_context` / `add_to_shared_context` / `join_shared_context` / `leave_shared_context`	Shared context
`end_session`	Session 終了

詳細: creo-memories/reference/mcp-tools.md

Future API (v0.24+ 予定)

70 tool → 6 core verbs + 5 named conveniences = 11 tools へリデザイン提案中:

Core: read / write / remove / query / transform / subscribe
Named: remember / recall / annotate / complete_with_context / record_work_log / end_session

詳細: creo-memories/reference/api-redesign.md

Setup / Authentication

Plugin install (/install chronista-club/claude-plugin-creo-memories)
初回利用時 OAuth (Auth0) prompt — Google / GitHub login
以後 auto

API Key (programmatic):

generate_api_key で発行
Authorization: Bearer <key> header で使用

Requirements

Claude Code
Creo Memories account (free tier)

Links

	URL
MCP Endpoint	`https://mcp.creo-memories.in`
Web Viewer	`https://creo-memories.in`
GitHub	`https://github.com/chronista-club/claude-plugin-creo-memories`

License

MIT