kiokuko-dsh 0.1.50 → 0.1.51

package/README.ja.md

@@ -2,15 +2,17 @@
 
 [English](README.md) | 日本語 | [简体中文](README.zh-CN.md) | [한국어](README.ko.md)
 
-[DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) に、プロジェクトの記憶・作業計画・検証支援を追加するプラグインです。
-OrcaReplay でモデル・ツールの動作を記録し、内容の確認と HTML 出力もできます。
+[DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness) に、記憶・計画・検証・動作記録を追加するプラグインです。
 
+## 機能
 
-新しい変更作業では、通常実行か役小角(enno-oduno)を選べます。役小角ではおすすめテンプレート、またはDSH設定済みモデルから役割ごとの構成を選択します。[モデル選択と接続上の制約](docs/model-selection.ja.md)
-
-`/deep-planning <問題>` は、四つの役割による読み取り専用の調査・分析・計画作成です。入力を保持し、一時停止・復旧に対応します。推定予算に達した場合も部分回答を返します。[操作・予算・復旧の仕様](docs/deep-planning.md)
-
-DeepSeek・Kimi・GLM・Qwen・HY/Hunyuan・MiMo・MiniMaxのAgentには、日本語を自然に整える同梱Skillを自動で渡します。[適用条件とモデル判定](docs/japanese-output.md)
+- **実行方式** — 通常実行と役小角(enno-oduno)を選べます。[詳細](docs/model-selection.ja.md)
+- **プロジェクト記憶** — 作業で得た知識を後のタスクで検索できます。[基本概念](docs/concepts.ja.md)
+- **Deep Planning** — 四つの役割で読み取り専用の調査・計画を行います。[使い方](docs/deep-planning.md)
+- **Continuity** — 既定では無効です。`continuity.mode: active` にすると、最近の実行記録を短くまとめてモデルへ渡します。[設定方法](docs/continuity.ja.md)
+- **記憶学習** — 完了した作業から再利用できる episode・教訓候補を作ります。[設定](docs/memory-evolution.md)
+- **OrcaReplay** — モデル・ツールの動作を記録し、HTML に出力できます。[設定とコマンド](docs/orca-recording.md)
+- **日本語出力** — 対応モデルへ自然な日本語を書く同梱 Skill を渡します。[詳細](docs/japanese-output.md)
 
 ## 導入と使い方
 
@@ -27,17 +29,6 @@ pnpm dsh web
 起動後は普通に依頼を入力します。Kiokuko 専用の setup 操作は不要です。
 GitHub・ローカルからの導入は [プラグインガイド](docs/dsh-plugin.md) を参照してください。
 
-OrcaReplay の機能設定は自動導入され、**手動設定は不要**です。各チャットは確認なしで詳細ログを記録します。
-記録すると、モデル応答やツール実行結果が作業プロジェクトの `.orca/runs/` に保存されます。判断はセッションごとに保持されるため、`/kioku-orca stop` したチャットは記録されないままです。毎回確認したい場合は `orca.askOnStart: true` を設定します。
-
-- `/kioku-orca start`: 手動で記録を開始、または停止後に再開します。そのチャットを停止していなければ実行不要です。過去の動作は記録されません。
-- `/kioku-orca status`: 記録状態・保存先・次の操作を短く表示します。診断用の詳細情報は `/kioku-orca status --json` で確認できます。
-
-`/kioku-orca stop` でログを確定後、`list` で run ID を確認し、`show <run ID>` で内容を表示、`export <run ID>` で HTML を出力できます（いずれも `/kioku-orca` に続けて入力）。
-無効にする場合は `orca.enabled: false` を設定して再読み込みします。詳細は [記録設定とコマンド](docs/orca-recording.md) を参照してください。
-
-終了した作業から episode と教訓候補を生成します。初期設定の `active` では、根拠が有効な候補を次のセッションの検索・自動注入に使います。候補生成だけにする場合は `observe`、停止する場合は `off` を指定します。状態は `/kioku-evolution status`、詳細は [記憶学習の設定・検証](docs/memory-evolution.md) を参照してください。
-
 ## 更新
 
 作業中のタスクを終えて DSH を停止してから更新します。npm 版の Kiokuko 本体の更新:
@@ -47,6 +38,17 @@ pnpm dsh plugin --profile web update kiokuko-dsh --latest
 pnpm dsh web
 ```
 
+Kiokuko は更新頻度が高く、pnpm 11 では公開後24時間未満のバージョンを
+[`minimumReleaseAge`](https://pnpm.io/settings/dependency-resolution#minimumreleaseage)
+により既定で選びません。`update --latest` を実行しても旧版のままになる場合は、
+`~/.dsh/profiles/web/pnpm-workspace.yaml`（`pnpm-lock.yaml` ではありません）の
+`minimumReleaseAgeExclude` に次を追加し、更新コマンドを再実行します。
+
+```yaml
+minimumReleaseAgeExclude:
+  - kiokuko-dsh
+```
+
 起動時に、日本語出力を含む全8 Skillと参照ファイルを `~/.agents/skills/` へ同期します。不足分を作成し、管理対象の旧版を更新します。非管理ファイルは上書きしません。他のエージェントではSkillカタログの再読み込みが必要です。
 
 起動時には、起動ディレクトリの `AGENTS.md` にある既存の Kiokuko 管理ブロックも更新し、その外側の指示は保持します。パッケージ更新後は DSH を再起動してください。別の作業ディレクトリや配備コピーの確認・修復は[セットアップ手順](docs/dsh-plugin.md#what-setup-updates-and-when)を参照してください。`kiokuko use` は DSH のセットアップコマンドではありません。

package/README.ko.md

@@ -40,6 +40,17 @@ pnpm dsh plugin --profile web update kiokuko-dsh --latest
 pnpm dsh web
 ```
 
+Kiokuko는 자주 업데이트되며, pnpm 11은 기본적으로
+[`minimumReleaseAge`](https://pnpm.io/settings/dependency-resolution#minimumreleaseage)를
+적용해 공개 후 24시간이 지나지 않은 버전을 선택하지 않습니다. `update --latest`를 실행해도
+이전 버전이 유지되면 `~/.dsh/profiles/web/pnpm-workspace.yaml`(`pnpm-lock.yaml`이 아님)의
+`minimumReleaseAgeExclude`에 다음 설정을 추가한 뒤 업데이트 명령을 다시 실행하세요.
+
+```yaml
+minimumReleaseAgeExclude:
+  - kiokuko-dsh
+```
+
 시작 시 일본어 출력 Skill을 포함한 번들 Skill 8개와 참조 파일을 `~/.agents/skills/`에 동기화합니다. 누락된 파일을 만들고 관리 대상 복사본을 업데이트하며, 비관리 파일은 덮어쓰지 않습니다. 다른 에이전트에서는 Skill 카탈로그를 다시 로드해야 합니다.
 
 시작 시 시작 디렉터리의 `AGENTS.md`에 있는 기존 Kiokuko 관리 블록도 갱신하며, 블록 밖의 지침은 보존합니다. 패키지 업데이트 후 DSH를 다시 시작하세요. 다른 작업 디렉터리나 오래된 복사본의 확인 및 복구는 [설정 절차](docs/dsh-plugin.md#what-setup-updates-and-when)를 참고하세요. `kiokuko use`는 DSH 설정 명령이 아닙니다.

package/README.md

@@ -2,16 +2,18 @@
 # Kiokuko(記憶庫) DeepSeek Harness Plugin
 [日本語](README.ja.md) | [简体中文](README.zh-CN.md) | [한국어](README.ko.md)
 
-Kiokuko adds project memory, planning, and verification support to
+Kiokuko adds memory, planning, verification, and observability to
 [DeepSeek Harness](https://github.com/deepseek-ai/deepseek-harness).
-OrcaReplay records model/tool activity for inspection and HTML export.
 
+## Features
 
-For new coding tasks, choose normal execution or 役小角(enno-oduno). Enno offers model templates and role assignments from configured DSH models. [Model selection and connection limits](docs/model-selection.md).
-
-Use `/deep-planning <problem>` for bounded, read-only investigation and planning with four agent roles. It preserves inputs, supports pause/recovery, and returns partial answers when its estimated budget is exhausted. [Commands, budgets and recovery](docs/deep-planning.md).
-
-DeepSeek, Kimi, GLM, Qwen, HY/Hunyuan, MiMo and MiniMax agents automatically receive the bundled Japanese writing Skill. [Application rules and model identification](docs/japanese-output.md).
+- **Execution modes** — Choose normal execution or 役小角(enno-oduno) for role-based planning and verification. [Details](docs/model-selection.md)
+- **Project memory** — Retrieve useful project knowledge for later work. [Concepts](docs/concepts.md)
+- **Deep planning** — Run bounded, read-only investigation with four roles. [Usage](docs/deep-planning.md)
+- **Continuity** — Disabled by default. Set `continuity.mode: active` to summarize recent execution evidence for the model. [Setup](docs/continuity.md)
+- **Memory evolution** — Turn completed work into reusable episode and lesson candidates. [Settings](docs/memory-evolution.md)
+- **OrcaReplay** — Record model/tool activity and export it as HTML. [Settings and commands](docs/orca-recording.md)
+- **Japanese output** — Give supported models a bundled Skill for natural Japanese. [Details](docs/japanese-output.md)
 
 ## Install and use
 
@@ -28,15 +30,6 @@ If you use a globally installed `dsh` CLI, omit `pnpm` from the commands.
 Enter your task normally; no Kiokuko setup command is needed.
 For GitHub/local installation, see the [plugin guide](docs/dsh-plugin.md).
 
-OrcaReplay is configured automatically; **no manual setup is required**. Every chat records detailed logs without asking.
-Recording saves model responses and tool results to `.orca/runs/` in the session workspace. A decision is remembered per session, so `/kioku-orca stop` keeps that chat unrecorded; set `orca.askOnStart: true` to ask at the start of each chat instead.
-
-- `/kioku-orca start`: start recording manually or resume after stopping. Not needed unless this chat was stopped earlier; past activity is not captured.
-- `/kioku-orca status`: briefly show the recording state, storage location, and next action. Use `/kioku-orca status --json` for diagnostic details.
-
-Use `/kioku-orca stop` to finalize the log, `list` to find its run ID, `show <run ID>` to inspect it, and `export <run ID>` to create HTML (all under `/kioku-orca`).
-To disable recording, set `orca.enabled: false` and reload. See [recording settings and commands](docs/orca-recording.md).
-
 ## Update
 
 Finish active tasks and stop DSH before updating. Update the npm-installed plugin:
@@ -46,6 +39,17 @@ pnpm dsh plugin --profile web update kiokuko-dsh --latest
 pnpm dsh web
 ```
 
+Kiokuko releases frequently, and pnpm 11 applies a 24-hour
+[`minimumReleaseAge`](https://pnpm.io/settings/dependency-resolution#minimumreleaseage)
+by default. If `update --latest` keeps the previous version, add Kiokuko to
+`minimumReleaseAgeExclude` in `~/.dsh/profiles/web/pnpm-workspace.yaml` (not
+`pnpm-lock.yaml`), then run the update command again:
+
+```yaml
+minimumReleaseAgeExclude:
+  - kiokuko-dsh
+```
+
 At startup, all eight bundled Skills (including Japanese output) and their references are synchronized to `~/.agents/skills/`. Missing files are created and managed copies are updated; unmanaged files are preserved. Other agents must reload their Skill catalog.
 
 Startup also refreshes an existing Kiokuko managed block in the startup directory’s `AGENTS.md`, preserving your other instructions. Restart DSH after updating the package. See [setup and verification](docs/dsh-plugin.md#what-setup-updates-and-when) for another workspace or stale copies; `kiokuko use` is not the DSH setup command.

package/README.zh-CN.md

@@ -40,6 +40,17 @@ pnpm dsh plugin --profile web update kiokuko-dsh --latest
 pnpm dsh web
 ```
 
+Kiokuko 更新频繁，而 pnpm 11 默认通过
+[`minimumReleaseAge`](https://pnpm.io/settings/dependency-resolution#minimumreleaseage)
+延迟选择发布不足24小时的版本。如果执行 `update --latest` 后仍停留在旧版本，请在
+`~/.dsh/profiles/web/pnpm-workspace.yaml`（不是 `pnpm-lock.yaml`）的
+`minimumReleaseAgeExclude` 中添加以下设置，然后重新执行更新命令：
+
+```yaml
+minimumReleaseAgeExclude:
+  - kiokuko-dsh
+```
+
 启动时会将包括日语输出在内的全部8个内置 Skill 及其引用文件同步到 `~/.agents/skills/`，创建缺失文件并更新受管理的副本，不覆盖非受管理文件。其他代理需要重新加载 Skill 目录。
 
 启动时也会更新启动目录中 `AGENTS.md` 的现有 Kiokuko 管理块，保留块外的指令。更新包后请重启 DSH。其他工作目录或旧副本的检查与修复请参阅[设置步骤](docs/dsh-plugin.md#what-setup-updates-and-when)。`kiokuko use` 不是 DSH 的设置命令。

package/dist/client.cjs

@@ -390,7 +390,10 @@ window.__ModuleLoader__.load({
         const question = pending.questions[0];
         const inputKind = questionInputKind(question);
         const mac = /Mac|iPhone|iPad|iPod/u.test(globalThis.navigator?.platform ?? '');
-        const shortcutModifier = mac ? 'Cmd' : 'Ctrl';
+        const userAgent = globalThis.navigator?.userAgent ?? '';
+        const safari = mac && /Safari\//u.test(userAgent) && !/(?:Chrome|Chromium|CriOS|Edg|OPR)\//u.test(userAgent);
+        const controlShortcut = !mac || safari;
+        const shortcutModifier = controlShortcut ? 'Ctrl' : 'Cmd';
         const [draft, setDraft] = useState(() => intakeDrafts.get(pending) ?? { selected: null, custom: '' });
         const [busy, setBusy] = useState(false);
         const [error, setError] = useState('');
@@ -453,7 +456,7 @@ window.__ModuleLoader__.load({
         const selectShortcut = (event) => {
             if (busy || inFlight.current || event.repeat || event.isComposing || event.keyCode === 229
                 || event.nativeEvent?.isComposing || event.nativeEvent?.keyCode === 229 || event.altKey || event.shiftKey
-                || !(mac ? event.metaKey && !event.ctrlKey : event.ctrlKey && !event.metaKey))
+                || !(controlShortcut ? event.ctrlKey && !event.metaKey : event.metaKey && !event.ctrlKey))
                 return false;
             const digit = /^(?:Digit|Numpad)([1-9])$/u.exec(event.code ?? '')?.[1] ?? event.key;
             if (!/^[1-9]$/u.test(digit) || Number(digit) > question.options.length)
@@ -477,7 +480,7 @@ window.__ModuleLoader__.load({
             };
             owner.addEventListener('keydown', listener, true);
             return () => owner.removeEventListener('keydown', listener, true);
-        }, [pending, busy, mac]);
+        }, [pending, busy, controlShortcut]);
         const keyDown = (event) => {
             if (selectShortcut(event))
                 return;
@@ -522,7 +525,7 @@ window.__ModuleLoader__.load({
                                 key: option.label, type: 'button', className: 'kiokuko-intake-option', disabled: busy,
                                 ref: (element) => { optionElements.current[index] = element; },
                                 'aria-pressed': draft.selected === index,
-                                ...(index < 9 ? { 'aria-keyshortcuts': `${question.options.length <= 9 ? `${index + 1} ` : ''}${mac ? 'Meta' : 'Control'}+${index + 1}` } : {}),
+                                ...(index < 9 ? { 'aria-keyshortcuts': `${question.options.length <= 9 ? `${index + 1} ` : ''}${controlShortcut ? 'Control' : 'Meta'}+${index + 1}` } : {}),
                                 onClick: () => update({ selected: index, custom: '' }),
                                 onKeyDown: (event) => {
                                     if (event.key !== 'Enter' || event.repeat || event.shiftKey || event.ctrlKey || event.altKey || event.metaKey || event.nativeEvent?.isComposing || event.nativeEvent?.keyCode === 229 || (intakeDrafts.get(pending) ?? draft).selected !== index)

package/dist/client.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"client.d.ts","sourceRoot":"","sources":["../src/client.ts"],"names":[],"mappings":"AAKA,UAAU,wBAAwB;IAChC,kBAAkB,CAAC,EAAE,CAAC,OAAO,EAAE,OAAO,KAAK,OAAO,CAAC;QAAE,cAAc,IAAI,OAAO,CAAC,cAAc,CAAC,UAAU,CAAC,CAAC,CAAA;KAAE,CAAC,CAAA;IAC7G,QAAQ,EAAE;QAAE,MAAM,CAAC,GAAG,EAAE,MAAM,GAAG,IAAI,CAAA;KAAE,CAAA;CACxC;AAED,MAAM,WAAW,yBAAyB;IACxC,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAA;IACzB,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;IAC1B,QAAQ,CAAC,KAAK,CAAC,EAAE,OAAO,UAAU,CAAC,KAAK,CAAA;IACxC,QAAQ,CAAC,MAAM,CAAC,EAAE,WAAW,CAAA;IAC7B,QAAQ,CAAC,MAAM,CAAC,EAAE,wBAAwB,CAAA;CAC3C;AAYD,UAAU,gBAAgB;IACxB,QAAQ,CAAC,cAAc,EAAE;QAAE,QAAQ,CAAC,MAAM,EAAE;YAAE,QAAQ,CAAC,UAAU,EAAE,OAAO,GAAG,OAAO,CAAA;SAAE,CAAA;KAAE,CAAA;IACxF,QAAQ,CAAC,MAAM,EAAE;QACf,QAAQ,CAAC,SAAS,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,GAAG,OAAO,CAAA;KAC3F,CAAA;IACD,QAAQ,CAAC,KAAK,EAAE;QACd,MAAM,CAAC,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,OAAO,GAAG,OAAO,CAAA;QACtD,QAAQ,CACN,UAAU,EAAE;YACV,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;YACrB,QAAQ,CAAC,EAAE,CAAC,EAAE,MAAM,CAAA;YACpB,QAAQ,CAAC,GAAG,CAAC,EAAE,MAAM,CAAA;YACrB,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAA;YAC1B,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,KAAK,OAAO,CAAA;YAC7D,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAA;YACvB,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;SAChD,EACD,SAAS,EAAE,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,KAAK,OAAO,GACrD,OAAO,CAAA;KACX,CAAA;IACD,MAAM,CAAC,KAAK,EAAE,MAAM,IAAI,GAAG,CAAC,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,EAAE,KAAK,EAAE,MAAM,GAAG,OAAO,CAAA;IAChF,EAAE,CAAC,KAAK,EAAE,kBAAkB,EAAE,QAAQ,EAAE,CAAC,SAAS,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,MAAM,EAAE;QAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,GAAG,OAAO,CAAA;CACtI;AA0ED,+EAA+E;AAC/E,wBAAsB,qBAAqB,CAAC,OAAO,EAAE,yBAAyB,GAAG,OAAO,CAAC,UAAU,GAAG,WAAW,CAAC,CAqBjH;AAuHD,eAAO,MAAM,MAAM,gDAAiD,CAAA;AAsVpE,mEAAmE;AACnE,wBAAgB,KAAK,CAAC,GAAG,EAAE,gBAAgB,GAAG,IAAI,CAgCjD"}
+{"version":3,"file":"client.d.ts","sourceRoot":"","sources":["../src/client.ts"],"names":[],"mappings":"AAKA,UAAU,wBAAwB;IAChC,kBAAkB,CAAC,EAAE,CAAC,OAAO,EAAE,OAAO,KAAK,OAAO,CAAC;QAAE,cAAc,IAAI,OAAO,CAAC,cAAc,CAAC,UAAU,CAAC,CAAC,CAAA;KAAE,CAAC,CAAA;IAC7G,QAAQ,EAAE;QAAE,MAAM,CAAC,GAAG,EAAE,MAAM,GAAG,IAAI,CAAA;KAAE,CAAA;CACxC;AAED,MAAM,WAAW,yBAAyB;IACxC,QAAQ,CAAC,QAAQ,EAAE,MAAM,CAAA;IACzB,QAAQ,CAAC,SAAS,EAAE,MAAM,CAAA;IAC1B,QAAQ,CAAC,KAAK,CAAC,EAAE,OAAO,UAAU,CAAC,KAAK,CAAA;IACxC,QAAQ,CAAC,MAAM,CAAC,EAAE,WAAW,CAAA;IAC7B,QAAQ,CAAC,MAAM,CAAC,EAAE,wBAAwB,CAAA;CAC3C;AAYD,UAAU,gBAAgB;IACxB,QAAQ,CAAC,cAAc,EAAE;QAAE,QAAQ,CAAC,MAAM,EAAE;YAAE,QAAQ,CAAC,UAAU,EAAE,OAAO,GAAG,OAAO,CAAA;SAAE,CAAA;KAAE,CAAA;IACxF,QAAQ,CAAC,MAAM,EAAE;QACf,QAAQ,CAAC,SAAS,EAAE,MAAM,EAAE,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,GAAG,OAAO,CAAA;KAC3F,CAAA;IACD,QAAQ,CAAC,KAAK,EAAE;QACd,MAAM,CAAC,IAAI,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,OAAO,GAAG,OAAO,CAAA;QACtD,QAAQ,CACN,UAAU,EAAE;YACV,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;YACrB,QAAQ,CAAC,EAAE,CAAC,EAAE,MAAM,CAAA;YACpB,QAAQ,CAAC,GAAG,CAAC,EAAE,MAAM,CAAA;YACrB,QAAQ,CAAC,QAAQ,CAAC,EAAE,MAAM,CAAA;YAC1B,QAAQ,CAAC,MAAM,CAAC,EAAE,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,KAAK,OAAO,CAAA;YAC7D,QAAQ,CAAC,MAAM,EAAE,MAAM,CAAA;YACvB,QAAQ,CAAC,MAAM,CAAC,EAAE,MAAM,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;SAChD,EACD,SAAS,EAAE,CAAC,KAAK,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,KAAK,OAAO,GACrD,OAAO,CAAA;KACX,CAAA;IACD,MAAM,CAAC,KAAK,EAAE,MAAM,IAAI,GAAG,CAAC,MAAM,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC,EAAE,KAAK,EAAE,MAAM,GAAG,OAAO,CAAA;IAChF,EAAE,CAAC,KAAK,EAAE,kBAAkB,EAAE,QAAQ,EAAE,CAAC,SAAS,EAAE,MAAM,EAAE,WAAW,EAAE,MAAM,EAAE,MAAM,EAAE;QAAE,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;KAAE,KAAK,IAAI,GAAG,OAAO,CAAA;CACtI;AA0ED,+EAA+E;AAC/E,wBAAsB,qBAAqB,CAAC,OAAO,EAAE,yBAAyB,GAAG,OAAO,CAAC,UAAU,GAAG,WAAW,CAAC,CAqBjH;AAuHD,eAAO,MAAM,MAAM,gDAAiD,CAAA;AAyVpE,mEAAmE;AACnE,wBAAgB,KAAK,CAAC,GAAG,EAAE,gBAAgB,GAAG,IAAI,CAgCjD"}

package/docs/continuity.ja.md

@@ -0,0 +1,31 @@
+# Continuity
+
+## 機能
+
+最近の実行記録を短くまとめ、`kiokuko:execution` に表示します。
+
+- `off`: 無効（既定値）
+- `shadow`: 要約を作るが、モデルへの入力は変えない
+- `active`: 最近の証拠表示を要約に置き換える
+
+## メリット
+
+何を読んだか、次に何を確認すべきかをモデルが把握しやすくなります。
+要約のサイズと件数には上限があるため、コンテキストの増加も抑えられます。
+
+## 設定方法
+
+`~/.dsh/profiles/web/cordis.patch.yml` に次を追加します。
+
+```yaml
+- id: kiokuko-dsh
+  config:
+    continuity:
+      mode: active
+      maxSupplementBytes: 4096
+      maxItems: 12
+```
+
+同じ行にほかの `config` がある場合は、その設定を残したまま `continuity` を追加してください。
+`web` プロファイルでは自動で反映されます。設定内容は
+`dsh --profile web --dump-config` で確認できます。

package/docs/continuity.md

@@ -1,193 +1,32 @@
-# Continuity projection
+# Continuity
 
-Continuity is an opt-in, disposable view of existing execution records. It does
-not save model hidden state, train a model, add memory tables, issue model
-requests, or advance a run. Deep planning keeps its existing prompts and state.
+## What it does
 
-## Configuration
+Continuity adds a short summary of recent execution evidence to
+`kiokuko:execution`.
 
-Set the Kiokuko plugin configuration and reload it through the existing DSH
-plugin lifecycle:
+- `off`: disabled (default)
+- `shadow`: builds the summary without changing the model input
+- `active`: replaces the recent-evidence section with the summary
 
-```yaml
-continuity:
-  mode: off                 # off | shadow | active
-  maxSupplementBytes: 4096   # integer, 512..8192 UTF-8 bytes
-  maxItems: 12               # integer, 1..24
-efficiency:
-  observe: true              # optional bounded in-memory numeric observations
-```
-
-`off` leaves the existing evidence presentation in place and does no continuity
-source loading or view construction. `shadow` constructs the proposed display
-but sends the same text as `off`. `active` replaces the recent-evidence portion
-of `kiokuko:execution`; it preserves the existing frame, constraints, exploration
-notice, terminal notice and formal Enno directive. It does not copy memory or
-the directive's criteria and next action into another permanent snapshot.
-
-Whole items are selected within the configured byte and item limits; at most
-three next-check items can be displayed. The footer always states coverage and
-omitted item count. The byte budget applies to the supplement, not the original
-request, mandatory conditions or directive. `complete` means that the selected
-adapter fields were available, not that the task succeeded or a whole document
-was read. The evidence window remains bounded to the existing 32 records; when
-that window is full, coverage is conservatively `partial` even if all displayed
-items fit. Its footer cannot count unknown records outside the loaded window.
-
-## Ownership and delivery
-
-The host resolves the native session/run binding, updates intake and the
-execution frame, and supplies committed evidence. The adapter and renderer are
-deterministic functions with no I/O. Source digests include the owner, frame
-content and generation; Enno additionally includes contract/mutation revision,
-route, directive and next action. A separate body digest describes display
-content. Neither digest is proof of correctness or authorization.
-
-The normal adapter ranks incomplete acquisition/presentation first, exact
-explicit read targets second, and recent observations next. All acquisition
-records are historical. A previous `presentation: full` is never asserted to be
-full in the current request. The final stream seam observes actual native tool
-results; the short supplement itself is not a full presentation of that result.
-
-The Enno adapter accepts only the bound run, native session, workspace, contract,
-route, role and current WorkUnit. It includes reports for that WorkUnit and its
-declared dependencies, explicitly labeled `model-report / unknown`. It does not
-promote `outcome: completed` into test success. Final verifier records remain
-`unknown` with respect to current repository freshness. The existing Enno service
-still owns the freshness and authorization checks. Raw verifier output is not
-copied into the supplement.
-
-The current integration needs one additional existing-service Enno snapshot read
-per nonterminal, unpaused model request in `shadow` or `active`. It uses the
-service's existing repository/verification checks; it does not run verifiers.
-This cost is not claimed as an optimization. Normal execution adds no DB read.
-Source mismatch or optional rendering failure degrades the supplement without
-changing the existing directive, tool outcome or authorization checks.
-
-Both native snapshot projection paths use retained surface events. An old
-snapshot remaining only in append-only history cannot suppress re-delivery or
-revive another plugin's compacted section. The current assembly's other sections
-are preserved. Without a direct surface API, the existing append/replace replay
-helper reconstructs retained events; invalid history cannot establish delivery.
-An uncommitted projection attempt is never evidence of delivery. No extra
-delivery cache or native session-event type is introduced.
-
-`efficiency.observe` exposes bounded `snapshot().continuity` entries containing
-only mode, bytes, item/omission counts, coverage and copies found in the final
-request. No source text, paths, credentials or source digest is recorded there.
-Observer errors leave `next()` and the original stream outcome unchanged.
-
-## Disable and recover
-
-Set `continuity.mode: off` and reload the plugin. The next request rebuilds the
-existing execution section, replacing the active supplement. The original
-conversation, conditions, evidence and Enno state remain. Resume uses the
-existing run/session route and lease rules; it neither replays tools nor moves
-state to a different run. No schema migration or rollback migration is needed.
-No unsupported new cross-session resume capability is added.
-
-## Verification and measurements
+## Benefits
 
-The baseline is commit `4094b061c6f713c6cde550b4b4cebe4bd8d715dd` (`0.1.49`).
-The starting local `0.1.14` checkout lacked the required modules; it was updated
-to the plan's baseline before implementation. The lockfile hashes and one
-scripted native lifecycle per mode are recorded in
-[`baseline.json`](../tests/fixtures/continuity/baseline.json).
+The model can quickly see what was already read and what still needs checking.
+The summary is limited by size and item count, so context growth stays bounded.
 
-Before the repair, the compaction counterexample returned zero snapshots where
-one current snapshot was required. The existing focused baseline had 16 passing
-tests. The added tests inspect the real DSH `0.1.5-rc.1` final model request after
-native surface replacement and same-run plugin reload, plus observer failures,
-late refreshes, mode changes, hostile text, source ownership, byte limits,
-dependency reports and verifier labels.
+## Setup
 
-All three native runs made 10 scripted model requests and five reads. The
-`shadow` request byte array matched `off` exactly. `active` increased bytes;
-this is not evidence of improved reasoning, speed or token efficiency. Mock
-usage is not real model usage. SQL counts, DB timings, expansion counts and
-provider tokens were not measured and remain `null`. PR-4 retrieval caching is
-therefore not implemented. No adoption or default-enable conclusion is drawn.
+Add the following to `~/.dsh/profiles/web/cordis.patch.yml`:
 
-```sh
-npm run typecheck
-npm run test:continuity
-# Include the real native-loop cases; missing runtime then fails instead of skipping.
-KIOKUKO_DSH_PACKAGE_ROOT=/path/to/pinned-dsh/node_modules \
-KIOKUKO_REQUIRE_DSH_NATIVE=1 npm run test:continuity
-```
-
-The runtime fixture is `tests/fixtures/dsh-runtime/package-lock.json`, as in CI.
-Verification used Node `v26.7.0`, npm `11.19.0` and DSH `0.1.5-rc.1` on macOS.
-Packaged install, Web bundle load, reload and uninstall were exercised with
-the existing CLI lifecycle checks in an isolated profile and home directory.
-
-Verification on 2026-09-14:
-
-| Check | Result |
-| --- | --- |
-| Typecheck, build, publint, pack closure/import checks | Passed |
-| Full suite with required pinned native runtime | 688 passed; one CLI-path-dependent test initially skipped |
-| Initially skipped bundle-install test with explicit `DSH_BIN` | Passed, no skip |
-| Final continuity suite (including real native requests) | 23 passed, no skips |
-| Real Enno native resume/verification/completion with active continuity | Passed |
-| Sample database initialization and resume | Passed |
-| Evolution regression suite | 33 passed |
-| Efficiency and Akinator synthetic evaluations | Completed; not model capability evidence |
-| Packaged CLI install/Web/reload/uninstall | Passed using the existing lifecycle portion separately from native tests |
-| Configured evaluation runner | Mock-fetch budget, privacy and output-reuse tests passed; no real provider used |
-
-Contracts were reviewed with `code.boundary.v1`, `code.protocol.v1` and
-`code.verification.v1`; the evaluation I/O also used `code.effects.v1`.
-The flow checked was config → native host → committed sources → execution
-snapshot → final request → numeric observation, including packaged activation
-and reload. No retrieved memory claims were supplied for this implementation.
-
-## Explicit model evaluation
-
-```sh
-npm run test:evaluation:continuity
-# No configuration: unmeasured; no network or model calls.
-npm run test:evaluation:continuity -- \
-  --config continuity-evaluation.local.json --output continuity-evaluation-results
-```
-
-Run these commands from a source checkout. The runner evaluates six fixed **context recovery probes**, not full repository
-tasks or a new autonomous loop. The fixture digest and exact JSON grading rules
-are fixed in `tests/fixtures/continuity/manifest.json` and `scenarios.json`.
-Each trial gets a disposable workspace, fixed input, model/settings and budget;
-it never opens the user's DB or sessions. A/B order is randomized by seed.
-Pairing and bootstrap intervals use scenarios, not individual turns. Small
-sample intervals do not establish the plan's end-to-end adoption criteria.
-The plan's larger, real-agent pilot remains a separate evaluation.
-
-Example configuration for a compatible chat-completions endpoint:
-
-```json
-{
-  "fixture": "continuity-v1",
-  "baseUrl": "http://127.0.0.1:8080/v1",
-  "model": "your-model",
-  "revision": "your-pinned-revision",
-  "temperature": 0,
-  "contextWindow": 32768,
-  "maxOutputTokens": 256,
-  "repetitions": 2,
-  "seed": 42,
-  "maxRequests": 24,
-  "maxTokens": 792576,
-  "maxDurationMs": 600000,
-  "allowRemote": false
-}
+```yaml
+- id: kiokuko-dsh
+  config:
+    continuity:
+      mode: active
+      maxSupplementBytes: 4096
+      maxItems: 12
 ```
 
-An optional `apiKeyEnv` names an environment variable; never put credentials in
-the file or URL. Optional `reasoningEffort` is sent explicitly. Remote endpoints
-require `allowRemote: true`; redirects are rejected. Each call reserves the
-declared context window plus maximum output against the total token budget.
-Reservations are conservative limits, not reported usage. Missing usage stays
-`null`; no retry is issued for uncertain failures. Cache state is uncontrolled
-and the declared immutable model revision is not independently verified.
-Existing reports are rejected before making calls. Reports contain bounded
-measurements and scores, not prompts, response text or authentication data.
-
-Real-model results are currently **unmeasured**. `off` remains the default.
+If the row already has other `config` values, keep them and add `continuity`
+alongside them. The `web` profile reloads the change automatically. Check the
+loaded configuration with `dsh --profile web --dump-config`.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "kiokuko-dsh",
-  "version": "0.1.50",
+  "version": "0.1.51",
   "description": "Kiokuko plugin for DeepSeek Harness",
   "type": "module",
   "license": "MIT",