Skip to content

A self-organizing Obsidian vault that gives AI coding agents persistent memory.

docs: dynamic command catalog, split doctrine, feature refresh across READMEs (#134)

Brenno Ferrari94b75f827f39
diff --git a/CLAUDE.md b/CLAUDE.md
index d985038..1eaa02c 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -20,28 +20,7 @@ This vault has [obsidian-skills](https://github.com/kepano/obsidian-skills) inst
### Custom Slash Commands
-Defined in `.claude/commands/`. See [[Skills]] for full documentation.
-
-| Command | Purpose |
-|---------|---------|
-| `/om-standup` | Morning kickoff -- load context, review yesterday, surface tasks, priorities |
-| `/om-dump` | Freeform capture -- dump anything, gets routed to the right notes |
-| `/om-wrap-up` | Full session review -- verify notes, indexes, links, suggest improvements |
-| `/om-humanize` | Voice-calibrated editing -- make notes sound like you, not AI |
-| `/om-weekly` | Weekly synthesis -- cross-session patterns, North Star alignment, uncaptured wins |
-| `/om-capture-1on1` | Capture 1:1 meeting transcript into structured vault note |
-| `/om-incident-capture` | Capture incident from Slack channels/DMs into structured vault notes |
-| `/om-slack-scan` | Deep scan Slack channels/DMs for evidence |
-| `/om-peer-scan` | Deep scan a peer's GitHub PRs for review prep |
-| `/om-review-brief` | Generate review brief (manager or peer version) |
-| `/om-self-review` | Write self-assessment for review tool -- projects, competencies, principles |
-| `/om-review-peer` | Write peer review -- projects, principles, performance summary |
-| `/om-vault-audit` | Audit indexes, links, orphans, stale context |
-| `/om-vault-upgrade` | Import content from an existing vault into this obsidian-mind instance |
-| `/om-prep-1on1` | Prep for an upcoming 1:1 -- load person context, open items, suggested agenda |
-| `/om-meeting` | Prep for any meeting by topic -- subject-forward briefing with open items and considerations |
-| `/om-intake` | Process meeting notes inbox -- classify and route to the right vault notes |
-| `/om-project-archive` | Move completed project from active/ to archive/, update indexes |
+Defined in `.claude/commands/`. Claude Code auto-surfaces every command with its description in the session's skills list — **that injected list is the live catalog**. Do not maintain a command table here (it drifts; the injected list can't — see Write-Correctness law 6). **Agents without that injection (Codex, Gemini, Cursor): read `brain/Skills.md`** — it carries the full command catalog plus usage docs and workflow sequences.
## Vault Structure
@@ -50,7 +29,7 @@ Defined in `.claude/commands/`. See [[Skills]] for full documentation.
| `Home.md` | **Vault entry point** -- embedded Base views, quick links | Open this first |
| `vault-manifest.json` | **Template metadata** -- version, infrastructure vs user content boundaries, frontmatter schemas, version fingerprints | Used by `/om-vault-upgrade` for migration |
| `CHANGELOG.md` | **Version history** -- tracks template releases (v1--v3.3) with what changed | Reference for upgrade paths |
-| `bases/` | **All Bases centralized** -- dynamic views for navigation | `Work Dashboard`, `Incidents`, `People Directory`, `1-1 History`, `Review Evidence`, `Competency Map`, `Templates` |
+| `bases/` | **All Bases centralized** -- dynamic views for navigation | `Work Dashboard` (incl. Stale Actives), `Recently Touched` (recency by real mtime — the answer to "what's most recent", not filename dates), `Incidents`, `People Directory`, `1-1 History`, `Review Evidence`, `Competency Map`, `Templates` |
| `work/` | Work notes index | `Index.md` (detailed MOC) |
| `work/active/` | **Current projects only** (1-3 files) | Move here when starting, move to archive when done |
| `work/archive/YYYY/` | Completed work organized by year | Grows over time |
@@ -69,8 +48,8 @@ Defined in `.claude/commands/`. See [[Skills]] for full documentation.
| `reference/` | Codebase knowledge, architecture maps | Flow docs, architecture docs |
| `thinking/` | Scratchpad for drafts and reasoning | Named `YYYY-MM-DD-topic.md` |
| `templates/` | Obsidian templates | `Work Note.md`, `Decision Record.md`, etc. |
-| `.claude/commands/` | 18 slash commands | See command table above |
-| `.claude/agents/` | 9 subagents | See subagents table below |
+| `.claude/commands/` | Slash commands (auto-surfaced in-session; catalog in `brain/Skills.md`) | One `.md` per command |
+| `.claude/agents/` | Subagents | See subagents table below |
| `.claude/scripts/` | Hook scripts | `session-start.ts`, `classify-message.ts`, `validate-write.ts`, `pre-compact.ts`, `stop-checklist.ts`, `charcount.ts` |
| `.claude/skills/` | Obsidian + QMD skills | Loaded automatically via Skill tool |
@@ -142,6 +121,7 @@ Use `thinking/` for drafts, reasoning, and analysis before writing final notes.
1. **Always use YAML frontmatter** with at minimum `date`, `description` (~150 chars), `tags`, and type-specific fields. Work notes and incidents also need `quarter` (e.g., `Q1-2026`). Incidents need `ticket`, `severity`, `role`.
2. **Use templates** from `templates/`. Fill `{{placeholders}}` with real values.
+2b. **Write fully, organize structurally — the vault tidies itself.** Size is a STRUCTURE signal, never a brevity signal: when a note crosses the ~25KB organization threshold (bytes, not lines — giant single-line entries hide in low line counts), the PostToolUse hook flags it at write time and the hygiene scan flags it at session boundaries. The response is always a SPLIT (domain notes, event-log satellites, or a cluster folder — content moved verbatim, one-liner index left behind, inbound links retargeted), never trimming content. The same hook flags write-time topic clusters — token overlap is blind, so judge genuine shared context before grouping. `/om-tidy` is the acting half. Exempt: `*Archive*` notes (bulk is their job).
3. **Place files correctly**:
- **Active** work notes, decisions, peer review prep -- `work/active/`
- **Completed** work notes -- `work/archive/YYYY/` (by year)
@@ -156,7 +136,8 @@ Use `thinking/` for drafts, reasoning, and analysis before writing final notes.
- Codebase knowledge -- `reference/`
- Drafts -- `thinking/`
- Vault root: `Home.md`, `CLAUDE.md`, `AGENTS.md`, `GEMINI.md`, `vault-manifest.json`, `CHANGELOG.md`, `CONTRIBUTING.md`, `README.md`, `LICENSE`, `.gitignore`. No user notes at root.
-4. **Name files descriptively.** Use the note title as filename.
+4. **Name files descriptively.** Use the note title as filename. **Date convention:** prefix **point-in-time** notes with a date — 1:1s (`<Person> YYYY-MM-DD.md`), dated captures, meeting exports. **Living and entity notes use a bare title** — work notes, person/team notes, competencies, `brain/`, `reference/`. *Why the split:* a creation-date prefix on a continuously-edited note **inverts the recency signal** (a living doc looks stale while being updated daily). For "what's most recent," use `bases/Recently Touched.base` (real modified time), never filename dates.
+5. **Group multi-note workstreams in a subfolder.** Once a workstream has more than one note, it gets a folder (`work/active/<Topic>/`; the archive mirrors the grouping). Folders are the lifecycle/context axis only — links remain the primary organization (wikilinks resolve by name across folders, so grouping never breaks links, QMD, or the graph). The hygiene scan and write-time cluster sensor flag candidates; `/om-project-archive` moves whole clusters.
### Note Types
@@ -171,6 +152,9 @@ Use `thinking/` for drafts, reasoning, and analysis before writing final notes.
| Team note | `org/teams/` | Team name | Members, Scope, Interactions |
| Competency | `perf/competencies/` | Competency name | Definition, level criteria, Evidence (via backlinks) |
| Brain note | `brain/` | Topic name | Topic-specific content |
+| Domain note | Beside its family index | `<Index> - <Domain>` (e.g. `Gotchas - Tooling`) | Substance lives here; the family index keeps a one-liner per entry. Born from monolith splits — never re-inline into the index |
+| Event-log satellite | Beside its core note | `<Core> — <Event> Log` or dated title | Chronological bulk offloaded from a person/project core note; the core links it. Dated entries, verbatim moves |
+| Archive note | Beside its live note | `<Live Name> Archive[ — <window>]` | Bulk by design, hygiene-exempt (name contains "Archive"). Verbatim zero-loss moves; live note keeps a one-liner index + link |
### Linking -- This Is Critical
diff --git a/README.ja.md b/README.ja.md
index 857dea7..7813e8d 100644
--- a/README.ja.md
+++ b/README.ja.md
@@ -167,7 +167,7 @@ qmd --index obsidian-mind embed # 多数の新規ノート後
**ボールトファーストのメモリ**がセッション間・マシン間でコンテキストを保持します。永続的な知識はすべて`brain/`のトピックノート(git管理、Obsidianで閲覧可能、リンク付き)に保存されます。Claude Codeの`MEMORY.md`(`~/.claude/`)はボールト内の場所を指す自動読み込みインデックスであり、ストレージそのものではありません。これにより、記憶はマシン変更後も生き残り、グラフの一部として機能します。
-**セッションには設計されたライフサイクルがあります。** `SessionStart`フックが自動的にNorth Starの目標、アクティブプロジェクト、最近の変更、未完了タスク、ボールト全体のファイル一覧を注入します — Claudeは白紙からではなく、コンテキストを持ってセッションを開始します。終了時に「wrap up」と言えば、Claudeが`/om-wrap-up`を実行 — ノートの検証、インデックスの更新、記録漏れの実績の発見を行います。285行の`CLAUDE.md`がその間のすべてを統制します:ファイルの配置場所、リンクの仕方、ノートの分割タイミング、意思決定やインシデントの扱い方。
+**セッションには設計されたライフサイクルがあります。** `SessionStart`フックが自動的にNorth Starの目標、アクティブプロジェクト、最近の変更、未完了タスク、ボールト全体のファイル一覧、ボールト衛生フラグを注入します — Claudeは白紙からではなく、コンテキストを持ってセッションを開始します。終了時に「wrap up」と言えば、Claudeが`/om-wrap-up`を実行 — ノートの検証、インデックスの更新、記録漏れの実績の発見を行います。`CLAUDE.md`がその間のすべてを統制します:ファイルの配置場所、リンクの仕方、ノートの分割タイミング、意思決定やインシデントの扱い方。
### 🔗 フック
@@ -175,11 +175,11 @@ qmd --index obsidian-mind embed # 多数の新規ノート後
| フック | タイミング | 内容 |
|------|------|------|
-| 🚀 SessionStart | 起動/再開時 | QMD再インデックス、North Star・アクティブプロジェクト・最近の変更・タスク・ファイル一覧を注入 |
+| 🚀 SessionStart | 起動/再開時 | QMD再インデックス+セルフヒール、North Starフォーカス・アクティブプロジェクト・最近の変更・タスク・ファイル一覧・ボールト衛生フラグを注入 — 最終行は注入サイズメーター |
| 💬 UserPromptSubmit | 全メッセージ | コンテンツを分類(意思決定、インシデント、実績、1on1、アーキテクチャ、人物、プロジェクト更新)してルーティングヒントを注入 |
-| ✍️ PostToolUse | `.md`書き込み後 | フロントマターの検証、ウィキリンクの確認 |
+| ✍️ PostToolUse | `.md`書き込み後 | フロントマター+ウィキリンクの検証、誤配置メモリファイルのブロック、肥大ノート(削るのではなく分割)と書き込み時トピッククラスタのフラグ |
| 💾 PreCompact | コンテキスト圧縮前 | セッション記録を`thinking/session-logs/`にバックアップ |
-| 🏁 Stop | セッション終了時 | チェックリスト:完了プロジェクトのアーカイブ、インデックス更新、孤立ノートの確認 |
+| 🏁 Stop | セッション終了時 | チェックリスト+具体的なドリフト検出(SessionStartと同じ衛生スキャン) |
> [!TIP]
> ただ話すだけ。フックがルーティングを処理します。
@@ -196,11 +196,11 @@ obsidian-mindはボールト全体をコンテキストに読み込み**ませ
| **トリガー** | PostToolUse検証 | `.md`書き込み後 | ~200トークン |
| **まれ** | ファイル全体の読み込み | 明示的に必要な場合のみ | 可変 |
-SessionStartは**軽量なコンテキスト**を読み込みます — 主要ファイルからの短い抜粋、ファイル名、git概要のみで、ノート全体は読み込みません。エージェントはQMDで意味検索してからファイルを読むため、関連する情報のみを取得します。分類フックはメッセージごとに軽量なNode呼び出し1回です。検証フックはMarkdown書き込み時のみ発火し、除外パスはスキップします。
+SessionStartは**軽量なコンテキスト**を読み込みます — 主要ファイルからの短い抜粋、ファイル名、git概要のみで、ノート全体は読み込みません。エージェントはQMDで意味検索してからファイルを読むため、関連する情報のみを取得します。分類フックはメッセージごとに軽量なNode呼び出し1回です。検証フックはMarkdown書き込み時のみ発火し、除外パスはスキップします。 ボールトが成長しても注入層を正直に保つ4つの仕組み: **ソース認識注入**(resume/compact時は揮発セクションのみ再注入 — 静的な大部分はすでに会話内にある)、全注入の最終行に付く**注入サイズメーター**、書き込みごとの**フック起動は1プロセスのみ**(QMDリフレッシュは検証フックに同乗)、**一覧の折りたたみ**(アーカイブは1行のカウント表示に)。
### 🌐 他のエージェントでの利用
-obsidian-mindはClaude Code、Codex CLI、Gemini CLIで動作します。`CLAUDE.md`のボールト規約、`.claude/scripts/`のフックスクリプト、`.claude/commands/`の18コマンドはすべてエージェント非依存です — 純粋なMarkdown、TypeScript、シェルでSDK依存はありません。
+obsidian-mindはClaude Code、Codex CLI、Gemini CLIで動作します。`CLAUDE.md`のボールト規約、`.claude/scripts/`のフックスクリプト、`.claude/commands/`のコマンドはすべてエージェント非依存です — 純粋なMarkdown、TypeScript、シェルでSDK依存はありません。
**Claude Code** — フルサポート。フック、コマンド、サブエージェント、メモリシステムがすべてそのまま動作します。
@@ -253,6 +253,7 @@ obsidian-mindはClaude Code、Codex CLI、Gemini CLIで動作します。`CLAUDE
| `/om-vault-audit` | インデックス、リンク、孤立ノート、古いコンテキストを監査 |
| `/om-vault-upgrade` | 既存ボールトからコンテンツをインポート — バージョン検出、分類、移行 |
| `/om-project-archive` | 完了プロジェクトをactive/からarchive/に移動し、インデックスを更新 |
+| `/om-tidy` | セルフメンテナンス — 衛生フラグに基づき、完了ノートのアーカイブ、クラスタのフォルダ化、肥大ノートの分割を実行。削除もコミットもしない |
---
@@ -299,7 +300,8 @@ obsidian-mindはClaude Code、Codex CLI、Gemini CLIで動作します。`CLAUDE
| Base | 表示内容 |
|------|---------|
-| Work Dashboard | 四半期でフィルタリングされ、ステータスでグループ化されたアクティブプロジェクト |
+| Work Dashboard | 四半期でフィルタリングされ、ステータスでグループ化されたアクティブプロジェクト — 14日以上更新のないアクティブノートを示す Stale Actives ビュー付き |
+| Recently Touched | 実際の更新時刻順の全ノート — 「最近何に取り組んだか」への正しい答え(生きたノートではファイル名の日付は当てにならない) |
| Incidents | 重要度と日付で並べ替えられた全インシデント |
| People Directory | `org/people/`の全員(役割、チーム付き) |
| 1:1 History | 人物と日付でソート可能な全1on1ノート |
diff --git a/README.ko.md b/README.ko.md
index 7b68017..d867605 100644
--- a/README.ko.md
+++ b/README.ko.md
@@ -167,7 +167,7 @@ qmd --index obsidian-mind embed # 많은 신규 노트 후
**볼트 우선 메모리**는 세션과 기기 간에 컨텍스트를 유지합니다. 모든 지속 지식은 `brain/` 토픽 노트에 있습니다(git 추적, Obsidian 브라우저 가능, 링크됨). Claude Code의 `MEMORY.md`(`~/.claude/`)는 볼트 위치를 가리키는 자동 로드 인덱스일 뿐 — 저장소 자체가 아닙니다. 즉, 메모리가 기기 변경에도 유지되며 그래프의 일부가 됩니다.
-**세션은 설계된 생명주기를 따릅니다.** `SessionStart` 훅이 North Star 목표, 활성 프로젝트, 최근 변경사항, 미완료 작업, 전체 볼트 파일 목록을 자동 주입합니다 — Claude는 빈 상태가 아니라 컨텍스트가 있는 상태로 모든 세션을 시작합니다. 마무리할 때 "wrap up"이라고 말하면 Claude가 `/om-wrap-up`을 실행합니다 — 노트 검증, 인덱스 업데이트, 누락된 성과 발견. 285줄의 `CLAUDE.md`가 그 사이의 모든 것을 관리합니다: 파일 배치 위치, 링크 방법, 노트를 분할해야 할 때, 의사결정과 인시던트 처리 방법.
+**세션은 설계된 생명주기를 따릅니다.** `SessionStart` 훅이 North Star 목표, 활성 프로젝트, 최근 변경사항, 미완료 작업, 전체 볼트 파일 목록, 볼트 위생 플래그를 자동 주입합니다 — Claude는 빈 상태가 아니라 컨텍스트가 있는 상태로 모든 세션을 시작합니다. 마무리할 때 "wrap up"이라고 말하면 Claude가 `/om-wrap-up`을 실행합니다 — 노트 검증, 인덱스 업데이트, 누락된 성과 발견. `CLAUDE.md`가 그 사이의 모든 것을 관리합니다: 파일 배치 위치, 링크 방법, 노트를 분할해야 할 때, 의사결정과 인시던트 처리 방법.
### 🔗 훅
@@ -175,11 +175,11 @@ qmd --index obsidian-mind embed # 많은 신규 노트 후
| 훅 | 시점 | 동작 |
|------|------|------|
-| 🚀 SessionStart | 시작/재개 시 | QMD 재인덱싱, North Star·활성 작업·최근 변경사항·작업·파일 목록 주입 |
+| 🚀 SessionStart | 시작/재개 시 | QMD 재인덱싱+셀프힐, North Star 포커스·활성 작업·최근 변경사항·작업·파일 목록·볼트 위생 플래그 주입 — 마지막 줄은 주입 크기 미터 |
| 💬 UserPromptSubmit | 매 메시지마다 | 콘텐츠 분류(의사결정, 인시던트, 성과, 1:1, 아키텍처, 인물, 프로젝트 업데이트) 및 라우팅 힌트 주입 |
-| ✍️ PostToolUse | `.md` 작성 후 | 프론트매터 유효성 검사, 위키링크 확인 |
+| ✍️ PostToolUse | `.md` 작성 후 | 프론트매터+위키링크 검증, 잘못 배치된 메모리 파일 차단, 비대 노트(줄이지 말고 분할)와 작성 시점 토픽 클러스터 플래그 |
| 💾 PreCompact | 컨텍스트 압축 전 | 세션 트랜스크립트를 `thinking/session-logs/`에 백업 |
-| 🏁 Stop | 세션 종료 시 | 체크리스트: 완료된 프로젝트 아카이브, 인덱스 업데이트, 고아 노트 확인 |
+| 🏁 Stop | 세션 종료 시 | 체크리스트+구체적인 드리프트 발견 (SessionStart와 동일한 위생 스캔) |
> [!TIP]
> 그냥 이야기하세요. 훅이 라우팅을 처리합니다.
@@ -196,11 +196,11 @@ obsidian-mind는 전체 볼트를 컨텍스트에 로드하지 **않습니다**.
| **트리거** | PostToolUse 검증 | `.md` 작성 후 | ~200 토큰 |
| **드물게** | 전체 파일 읽기 | 명시적으로 필요한 경우만 | 가변 |
-SessionStart는 **가벼운 컨텍스트**를 로드합니다 — 주요 파일의 짧은 발췌, 파일명, git 요약만 로드하며 전체 노트 내용은 읽지 않습니다. 에이전트는 QMD를 통해 의미 기반 검색 후 파일을 읽으므로 관련 정보만 가져옵니다. 분류 훅은 메시지당 가벼운 Node 호출 1회입니다. 검증 훅은 마크다운 쓰기 시에만 실행되며 제외된 경로는 건너뜁니다.
+SessionStart는 **가벼운 컨텍스트**를 로드합니다 — 주요 파일의 짧은 발췌, 파일명, git 요약만 로드하며 전체 노트 내용은 읽지 않습니다. 에이전트는 QMD를 통해 의미 기반 검색 후 파일을 읽으므로 관련 정보만 가져옵니다. 분류 훅은 메시지당 가벼운 Node 호출 1회입니다. 검증 훅은 마크다운 쓰기 시에만 실행되며 제외된 경로는 건너뜁니다. 볼트가 커져도 주입 계층을 정직하게 유지하는 네 가지 메커니즘: **소스 인식 주입** (resume/compact 시 휘발성 섹션만 재주입 — 정적인 대부분은 이미 대화에 있음), 모든 주입의 마지막 줄에 붙는 **주입 크기 미터**, 쓰기당 **훅 프로세스 1회** (QMD 리프레시가 검증 훅에 동승), **목록 접기** (아카이브는 한 줄 카운트로 접힘).
### 🌐 다른 에이전트와 사용하기
-obsidian-mind는 Claude Code, Codex CLI, Gemini CLI에서 작동합니다. `CLAUDE.md`의 볼트 규약, `.claude/scripts/`의 훅 스크립트, `.claude/commands/`의 18개 커맨드는 모두 에이전트 비의존적입니다 — 순수 Markdown, TypeScript, 셸이며 SDK 의존성이 없습니다.
+obsidian-mind는 Claude Code, Codex CLI, Gemini CLI에서 작동합니다. `CLAUDE.md`의 볼트 규약, `.claude/scripts/`의 훅 스크립트, `.claude/commands/`의 커맨드는 모두 에이전트 비의존적입니다 — 순수 Markdown, TypeScript, 셸이며 SDK 의존성이 없습니다.
**Claude Code** — 완전 지원. 훅, 커맨드, 서브에이전트, 메모리 시스템이 모두 기본으로 작동합니다.
@@ -253,6 +253,7 @@ obsidian-mind는 Claude Code, Codex CLI, Gemini CLI에서 작동합니다. `CLAU
| `/om-vault-audit` | 인덱스, 링크, 고아 노트, 오래된 컨텍스트 감사 |
| `/om-vault-upgrade` | 기존 볼트에서 콘텐츠 가져오기 — 버전 감지, 분류, 마이그레이션 |
| `/om-project-archive` | 완료된 프로젝트를 active/에서 archive/로 이동, 인덱스 업데이트 |
+| `/om-tidy` | 자체 유지보수 — 위생 플래그에 따라 완료 노트 아카이브, 클러스터 폴더화, 비대 노트 분할을 수행. 삭제도 커밋도 하지 않음 |
---
@@ -299,7 +300,8 @@ obsidian-mind는 Claude Code, Codex CLI, Gemini CLI에서 작동합니다. `CLAU
| Base | 표시 내용 |
|------|-------|
-| Work Dashboard | 분기별 필터링, 상태별 그룹화된 활성 프로젝트 |
+| Work Dashboard | 분기별 필터링, 상태별 그룹화된 활성 프로젝트 — 14일 이상 손대지 않은 활성 노트를 보여주는 Stale Actives 뷰 포함 |
+| Recently Touched | 실제 수정 시각 순의 모든 노트 — "최근 무엇을 작업했나"에 대한 정확한 답 (살아있는 노트에서 파일명 날짜는 믿을 수 없음) |
| Incidents | 심각도와 날짜순으로 정렬된 모든 인시던트 |
| People Directory | `org/people/`의 모든 인물과 역할, 팀 정보 |
| 1:1 History | 인물과 날짜로 정렬 가능한 모든 1:1 노트 |
diff --git a/README.md b/README.md
index 6266a6e..de2ddba 100644
--- a/README.md
+++ b/README.md
@@ -182,11 +182,11 @@ Five lifecycle hooks handle routing automatically:
| Hook | When | What |
|------|------|------|
-| 🚀 SessionStart | On startup/resume | QMD re-index, inject North Star, active work, recent changes, tasks, file listing |
+| 🚀 SessionStart | On startup/resume | QMD re-index + self-heal, inject North Star focus, active work, recent changes, tasks, file listing, vault-hygiene drift flags — ending with an injection-size meter |
| 💬 UserPromptSubmit | Every message | Classifies content (decision, incident, win, 1:1, architecture, person, project update) and injects routing hints |
-| ✍️ PostToolUse | After writing `.md` | Validates frontmatter, checks for wikilinks |
+| ✍️ PostToolUse | After writing `.md` | Validates frontmatter + wikilinks, blocks misplaced memory files, flags oversized notes (split, don't trim) and write-time topic clusters |
| 💾 PreCompact | Before context compaction | Backs up session transcript to `thinking/session-logs/` |
-| 🏁 Stop | End of session | Checklist: archive completed projects, update indexes, check orphans |
+| 🏁 Stop | End of session | Checklist + concrete drift findings (same hygiene scan as SessionStart) |
> [!TIP]
> You just talk. The hooks handle the routing.
@@ -203,7 +203,7 @@ obsidian-mind does **not** dump your entire vault into context. It uses tiered l
| **Triggered** | PostToolUse validation | After `.md` writes | ~200 tokens |
| **Rare** | Full file reads | Only when explicitly needed | Variable |
-SessionStart loads **lightweight context** — small excerpts from key files, filenames, and git summary — not full note contents. The agent queries by meaning via QMD before reading files, so it pulls only what's relevant. The classification hook is one lightweight Node call per message. The validation hook only fires on markdown writes and skips excluded paths.
+SessionStart loads **lightweight context** — small excerpts from key files, filenames, and git summary — not full note contents. Four mechanisms keep the eager layer honest as the vault grows: **source-aware injection** (resume/compact re-inject only volatile sections — the static bulk is already in-conversation), an **injection-size meter** as the last line of every injection (you always see what context costs), a **single hook spawn per write** (the QMD refresh rides the validation hook), and **listing collapse** (the archive folds to one count line). The agent queries by meaning via QMD before reading files, so it pulls only what's relevant. The classification hook is one lightweight Node call per message. The validation hook only fires on markdown writes and skips excluded paths.
### 🌐 Using with Other Agents
@@ -254,6 +254,7 @@ Defined in `.claude/commands/`. Run them in Claude Code, Codex CLI, or Gemini CL
| `/om-review-brief` | Generate a review brief (manager or peer version) |
| `/om-self-review` | Write your self-assessment for review season — projects, competencies, principles |
| `/om-review-peer` | Write a peer review — projects, principles, performance summary |
+| `/om-tidy` | Self-maintenance — acts on every hygiene flag: archive, group, split. Never deletes, never commits |
| `/om-vault-audit` | Audit indexes, links, orphans, stale context |
| `/om-vault-upgrade` | Import content from an existing vault — version detection, classification, migration |
| `/om-prep-1on1` | Prep for an upcoming 1:1 — load person context, open items, suggested agenda |
@@ -306,7 +307,8 @@ The `bases/` folder contains database views that query your notes' frontmatter p
| Base | Shows |
|------|-------|
-| Work Dashboard | Active projects filtered by quarter, grouped by status |
+| Work Dashboard | Active projects filtered by quarter, grouped by status — plus a Stale Actives view (active but untouched 14+ days) |
+| Recently Touched | Every note by real modified time — the correct answer to "what did I work on recently" (filename dates lie for living notes) |
| Incidents | All incidents sorted by severity and date |
| People Directory | Everyone in `org/people/` with role, team |
| 1:1 History | All 1:1 notes sortable by person and date |
diff --git a/README.zh-CN.md b/README.zh-CN.md
index a457eda..8691dea 100644
--- a/README.zh-CN.md
+++ b/README.zh-CN.md
@@ -167,7 +167,7 @@ qmd --index obsidian-mind embed # 大量新笔记后
**仓库优先的记忆** 使上下文能跨会话和设备持续存在。所有持久知识存储在 `brain/` 主题笔记中(git 跟踪、Obsidian 可浏览、相互链接)。Claude Code 的 `MEMORY.md`(`~/.claude/`)是一个自动加载的索引,指向仓库中的位置——本身不存储内容。这意味着记忆可以在更换设备后保留,并且是知识图谱的一部分。
-**会话有设计好的生命周期。** `SessionStart` 钩子自动注入你的北极星目标、活跃项目、最近变更、待办任务和完整的仓库文件列表——Claude 每次会话都带着上下文启动,而不是从空白开始。结束时,说 "wrap up",Claude 就会运行 `/om-wrap-up`——验证笔记、更新索引、发现未记录的成就。285 行的 `CLAUDE.md` 规范了中间的一切:文件归档位置、链接方式、何时拆分笔记、如何处理决策和事件。
+**会话有设计好的生命周期。** `SessionStart` 钩子自动注入你的北极星目标、活跃项目、最近变更、待办任务、完整的仓库文件列表和仓库卫生标志——Claude 每次会话都带着上下文启动,而不是从空白开始。结束时,说 "wrap up",Claude 就会运行 `/om-wrap-up`——验证笔记、更新索引、发现未记录的成就。`CLAUDE.md` 规范了中间的一切:文件归档位置、链接方式、何时拆分笔记、如何处理决策和事件。
### 🔗 钩子
@@ -175,11 +175,11 @@ qmd --index obsidian-mind embed # 大量新笔记后
| 钩子 | 触发时机 | 功能 |
|------|---------|------|
-| 🚀 SessionStart | 启动/恢复时 | QMD 重新索引,注入北极星目标、活跃工作、最近变更、任务、文件列表 |
+| 🚀 SessionStart | 启动/恢复时 | QMD 重新索引+自愈,注入北极星焦点、活跃工作、最近变更、任务、文件列表、仓库卫生标志 — 末行为注入大小计量 |
| 💬 UserPromptSubmit | 每条消息 | 对内容分类(决策、事件、成就、1:1、架构、人员、项目更新)并注入路由提示 |
-| ✍️ PostToolUse | 写入 `.md` 后 | 验证 frontmatter、检查 wikilinks |
+| ✍️ PostToolUse | 写入 `.md` 后 | 验证 frontmatter+wikilinks,拦截误放的记忆文件,标记过大笔记(拆分而非删减)与写入时的主题集群 |
| 💾 PreCompact | 上下文压缩前 | 将会话记录备份到 `thinking/session-logs/` |
-| 🏁 Stop | 会话结束时 | 检查清单:归档已完成项目、更新索引、检查孤立笔记 |
+| 🏁 Stop | 会话结束时 | 检查清单+具体漂移发现(与 SessionStart 相同的卫生扫描) |
> [!TIP]
> 你只需要正常对话。钩子会处理路由。
@@ -196,11 +196,11 @@ obsidian-mind **不会**将整个 vault 加载到上下文中。它使用分层
| **触发** | PostToolUse 验证 | `.md` 写入后 | ~200 tokens |
| **罕见** | 完整文件读取 | 仅在明确需要时 | 可变 |
-SessionStart 加载**轻量级上下文** — 关键文件的简短摘要、文件名和 git 摘要,而非完整笔记内容。Agent 通过 QMD 进行语义搜索后再读取文件,因此只获取相关内容。分类钩子每条消息仅执行一次轻量级 Node 调用。验证钩子仅在 markdown 写入时触发,跳过排除的路径。
+SessionStart 加载**轻量级上下文** — 关键文件的简短摘要、文件名和 git 摘要,而非完整笔记内容。Agent 通过 QMD 进行语义搜索后再读取文件,因此只获取相关内容。分类钩子每条消息仅执行一次轻量级 Node 调用。验证钩子仅在 markdown 写入时触发,跳过排除的路径。 四个机制让注入层随仓库增长仍保持诚实:**来源感知注入**(resume/compact 时只重新注入易变部分——静态大块已在会话中)、每次注入末行的**注入大小计量**、每次写入**仅一次钩子进程**(QMD 刷新搭载在验证钩子上)、**列表折叠**(归档折叠为一行计数)。
### 🌐 与其他 Agent 配合使用
-obsidian-mind 支持 Claude Code、Codex CLI 和 Gemini CLI。`CLAUDE.md` 中的 vault 规约、`.claude/scripts/` 中的钩子脚本、`.claude/commands/` 中的 18 个命令都是 Agent 无关的 — 纯 Markdown、TypeScript 和 Shell,无 SDK 依赖。
+obsidian-mind 支持 Claude Code、Codex CLI 和 Gemini CLI。`CLAUDE.md` 中的 vault 规约、`.claude/scripts/` 中的钩子脚本、`.claude/commands/` 中的命令都是 Agent 无关的 — 纯 Markdown、TypeScript 和 Shell,无 SDK 依赖。
**Claude Code** — 完整支持。钩子、命令、子代理和记忆系统全部开箱即用。
@@ -253,6 +253,7 @@ obsidian-mind 支持 Claude Code、Codex CLI 和 Gemini CLI。`CLAUDE.md` 中的
| `/om-vault-audit` | 审计索引、链接、孤立笔记、过时内容 |
| `/om-vault-upgrade` | 从现有仓库导入内容——版本检测、分类、迁移 |
| `/om-project-archive` | 将已完成的项目从 active/ 移至 archive/,更新索引 |
+| `/om-tidy` | 自维护 — 依据卫生标志归档已完成笔记、归组松散集群、拆分过大笔记。从不删除、从不提交 |
---
@@ -299,7 +300,8 @@ obsidian-mind 支持 Claude Code、Codex CLI 和 Gemini CLI。`CLAUDE.md` 中的
| Base | 显示内容 |
|------|---------|
-| Work Dashboard | 按季度筛选、按状态分组的活跃项目 |
+| Work Dashboard | 按季度筛选、按状态分组的活跃项目 — 附 Stale Actives 视图(14 天以上未动的活跃笔记) |
+| Recently Touched | 按真实修改时间排序的全部笔记 — 「最近在做什么」的正确答案(活笔记的文件名日期不可信) |
| Incidents | 按严重程度和日期排序的所有事件 |
| People Directory | `org/people/` 中所有人员的角色和团队 |
| 1:1 History | 可按人员和日期排序的所有 1:1 笔记 |
diff --git a/brain/Skills.md b/brain/Skills.md
index bddd742..4084603 100644
--- a/brain/Skills.md
+++ b/brain/Skills.md
@@ -8,7 +8,7 @@ tags:
# Skills
-Custom slash commands, subagents, and reusable workflows. Defined in `.claude/commands/` and `.claude/agents/`.
+Custom slash commands, subagents, and reusable workflows. Defined in `.claude/commands/` and `.claude/agents/`. **This note is the canonical command catalog** for agents that don't get the in-session skills injection (Codex, Gemini, Cursor) — keep it current when commands change; CLAUDE.md deliberately carries no command table.
## Slash Commands
@@ -53,7 +53,8 @@ Custom slash commands, subagents, and reusable workflows. Defined in `.claude/co
|---------|---------|
| `/om-vault-audit` | Deep structural audit — indexes, frontmatter, links, Bases, folder placement, stale context |
| `/om-vault-upgrade` | Import content from an existing vault — detects version, classifies notes, transforms frontmatter, rebuilds indexes |
-| `/om-project-archive` | Move completed project from `work/active/` to `work/archive/YYYY/`, update all indexes |
+| `/om-project-archive` | Move completed project from `work/active/` to `work/archive/YYYY/`, update all indexes — moves whole `active/<Topic>/` clusters intact |
+| `/om-tidy` | Self-maintenance pass — acts on every hygiene flag: archives completed work, groups clusters, splits oversized notes, reports open loops. Never deletes, never commits |
## Usage Notes