EoM

Taskr - Persistent Task Planning & Execution for AI Agents

AIエージェントv1.1.1
作者:EoM更新日 2026-05-11全 5 バージョン · 最新版の公開日 2026-03-23taskr@clawhub_echo-of-machines/taskr
タスク自動化マルチエージェント協調#Sync

AlphaMoatによると、2026-10時点でTaskr - Persistent Task Planning & Execution for AI Agentsのダウンロード数は8,378、スター数は3で、累計ダウンロード数で今回のランキング対象AIスキル10,914件中3,316位、AIエージェントでは1,632件中440位です。

一覧の初期表示は累計ダウンロード数順で、スター数順にも並べ替えできます。詳細ページの順位は累計ダウンロード数で計算しています。

ダウンロード数
8378
スター数
3
累計ダウンロード順位
#3,316
全体の上位 31% · AIエージェントの上位 27%
プラットフォームダウンロード数スター数コメント
ClawHub ↗60273-
SkillHub ↗83783—

プラットフォームごとに集計方法と集計時点が異なります。

このスキルでできること

タスク、背景メモ、状態をセッション間で保持し、別のエージェントが再開でき、ユーザーが計画と進捗を確認できます。担当の取得、完了、理由付きスキップ、実施済み作業の追記を扱います。接続には MCP_API_URL、MCP_USER_API_KEY、MCP_PROJECT_ID の設定が必要です。 このワークフローはリモートサービスまたは共有ワークスペースを使用します。非公開情報を送る前にアクセス範囲と保存期間を確認してください。

資料公開日 2026-10-27

できること

  • セッションをまたぐレビュー可能な計画と監査の記録を維持する

使う場面

  • 複数セッションや可視化した進捗、明確な引き継ぎが必要なとき

プロンプト例

スキルのドキュメントをもとに生成した例であり、実際の実行結果ではありません。

背景と完了条件を含む階層計画を準備し、実行前にレビュー用に示して。

月次の推移

ダウンロード数8378
4611663486572026-03 · ダウンロード数 48902026-05 · ダウンロード数 53682026-06 · ダウンロード数 54802026-07 · ダウンロード数 56512026-08 · ダウンロード数 58152026-09 · ダウンロード数 62522026-10 · ダウンロード数 837826-0326-0926-10
スター数3
2342026-03 · スター数 32026-05 · スター数 32026-06 · スター数 32026-07 · スター数 32026-08 · スター数 32026-09 · スター数 32026-10 · スター数 326-0326-0926-10

2026-09 から集計方法を変更し、同名スキルを作者ごとに分けて集計し、全スキルを対象にしました。破線は変更前の月で、変更後のデータとは直接比較せず、変更をまたぐ増減も計算していません。

月ダウンロード数前月比スター数
2026-108378—3
2026-096252—3
2026-085815+2.9%3
2026-075651+3.1%3
2026-065480+2.1%3
2026-055368—3
2026-034890—3

SKILL.md 原文

作者が公開した元のファイルをテキストのまま表示しています。AlphaMoat による編集や審査はしていません。リンクはクリックできません。コマンドは内容を確認してから実行してください。取得日 2026-10-23。

11 KB・ClawHub v1.1.1
---
name: taskr
description: "Persistent cloud task planning and execution for OpenClaw. Create hierarchical task plans that survive session resets, span multiple agents, and let users review and approve work before execution. Notes attach to tasks as audit trails. Use task_sync to retroactively record completed work. Homepage: https://taskr.one"
homepage: https://taskr.one
metadata: {"openclaw":{"emoji":"📋","requires":{"env":["MCP_API_URL","MCP_USER_API_KEY","MCP_PROJECT_ID"]},"primaryEnv":"MCP_USER_API_KEY"}}
---

# Taskr — Persistent Task Planning & Execution

Taskr gives OpenClaw agents persistent, structured task management that lives outside the chat session. Plans survive context resets, can be resumed by any agent on any machine, and are always visible to the user via the Taskr web app, VS Code extension, or mobile.
すべて表示(全 269 行)

**Six things Taskr does for OpenClaw:**
1. **Hierarchical planning** — break any work into a nested task hierarchy before touching a tool
2. **Persistent context** — tasks, notes, and status survive session resets and context compaction
3. **Cross-agent continuity** — any agent can pick up any task list from anywhere; `get_task` transfers ownership automatically
4. **Audit trail via notes** — notes attach directly to tasks; cause and effect stay together, not fragmented across chat history
5. **Retroactive sync** — `task_sync` records work done before Taskr was active, closing gaps in history
6. **Shared state** — statuses (`open`, `wip`, `done`, `skipped`) give both agent and user an unambiguous, real-time picture of progress

---

## When to Use Taskr

**Use Taskr when:**
- Work has 3+ steps or will take more than a few minutes
- Work spans multiple sessions or may be handed off to another agent
- User wants to monitor or approve progress remotely
- Resuming work from a previous session

**Skip Taskr for:**
- Single quick actions (<3 steps, <2 minutes)
- Pure information retrieval or simple questions
- User explicitly declines

**Proactive default:** For any substantial work, offer Taskr *before* starting:
> "I'll plan this in Taskr first — you can review the task breakdown before I start. Sound good?"

**Once Taskr is active, stay in Taskr.** Don't abandon tasks mid-workflow. Incomplete tasks in the dashboard are confusing and break the audit trail.

---

## The Core Loop

```
Plan → Create → Create CONTEXT note → Review with user → Execute → Document → Repeat
```

1. **Plan** — think through full scope; break into phases and subtasks before touching any tool
2. **Create** — `create_task` with `taskListTitle` to build the entire hierarchy in one call
3. **Create CONTEXT note** — always attach a CONTEXT note to the new task list (TL_) with background, goals, and any user preferences
4. **Review** — present the task plan to the user; get approval before executing anything
5. **Execute** — `get_task` → do the work → `update_task` status=done → repeat
6. **Document** — attach notes to tasks as you go; write PROGRESS notes at phase completions

**Single-task discipline:** Work on exactly one task at a time. `get_task` sets you as owner (wip). Complete or skip before moving on.

---

## Planning: Task Hierarchy Design

Use `create_task` with `taskListTitle` to create a new task list. Submit 1–100 tasks in a single call — always create the full hierarchy upfront so the user can review before execution begins.

**Hierarchy positions:**
- `"1"`, `"2"`, `"3"` — top-level phases (direct children of the task list)
- `"1.1"`, `"1.2"` — subtasks under phase 1
- `"1.1.1"` — deeper nesting (max 10 levels)
- Parents must be declared before their children in the same call

**Task types:** `setup`, `analysis`, `implementation`, `validation`, `testing`

**Keep tasks focused.** If a task feels too large to complete in one agent turn, break it into subtasks — at creation time or mid-execution via `create_task` + `parentId`. Subtasks prevent context overload and make progress visible at the right granularity.

**After creating a task list**, always immediately create a CONTEXT note on the task list:
```
create_note(type="CONTEXT", title="[Task List] — Context", body="...", taskId="TL_...")
```

---

## Executing: Working Through Tasks

### get_task
Call `get_task` to acquire your next task. It automatically sets status to `wip` and assigns ownership. Use `get_task` — do not manually set tasks to wip.

- Omit `task_list_id` to continue the current task list
- Pass `task_list_id=TL_...` to switch to a different task list
- Pass `include_context=true` for additional task context and notes
- Pass `bypass_task_id` only for emergency task-switching

### update_task — two modes

**Status mode** (one task):
```
update_task(taskId="TS_...", status="done", ruleContext="RU-PROC-001")
```

**Batch mode** (title/description updates only — status NOT allowed in batch):
```
update_task(tasks=[{taskId, title?, description?}, ...], ruleContext="RU-PROC-001")
```

### Task Statuses — Shared Checkpoints

Statuses are checkpoints visible to both agent and user in real time. Use them with discipline.

| Status | Meaning |
|--------|---------|
| `open` | Ready to be worked on |
| `wip` | Actively owned by an agent |
| `done` | Completed and verified |
| `skipped` | Deliberately not done — **must create a FINDING note explaining why** |

Completing the last child of a parent auto-marks the parent `done`. You don't need to mark parents manually.

### Handling Unexpected Work

When you discover work outside the current task's scope:

- **Blocks current task** → `create_task` (parentId=current task), complete subtask, resume
- **Unrelated** → finish current task first, then create a new task for the other work
- **Can wait** → document with a FINDING note, continue with current task

---

## Reviewing: get_task_hierarchy

Use `get_task_hierarchy` to browse and review tasks — not `get_task`. Never call `get_task` just to look at structure.

| Scope | What it returns |
|-------|----------------|
| `project` | All task lists. Use `depth=0` first for names only — avoids token bloat |
| `task_list` | Full nested hierarchy for one task list. Pass `taskId=TL_...` |
| `subtree` | A task and all its descendants. Pass `taskId=TS_...` |

Response path for `task_list` scope: `data.task_lists[0].tasks[0].children[]`

---

## Notes — Audit Trail Attached to Tasks

Notes attach to tasks. A finding discovered during task 2.3 lives on task 2.3 — not floating at the project level. Any agent resuming work later has full context exactly where they need it. This is what keeps information from becoming fragmented.

**Note types:**

| Type | When to use |
|------|-------------|
| `CONTEXT` | Background, goals, user preferences, decisions — create on TL_ immediately after task list creation |
| `FINDING` | Discoveries, issues, blockers, insights encountered during work |
| `PROGRESS` | Phase completions — **only for phase-level tasks** (tasks "1", "2", "3" — direct TL children). NOT for leaf tasks |
| `FILE_LIST` | Files created, modified, or deleted on the user's system |
| `OTHER` | Anything that doesn't fit the above |

**Note discipline:**
- Attach notes to the most specific relevant task (TS_ > TL_ > project-scoped)
- Update existing notes rather than creating duplicates — use `update_note`
- Search before creating: `search_notes` for prior context when resuming work
- `skipped` status always requires a FINDING note with the reason

**Two-step note discovery:**
1. `list_notes(taskId="TL_...", includeDescendants=true)` — get IDs and titles (`data.notes.notes[]`)
2. `get_note(noteId="NT_...")` — fetch full content when needed

---

## task_sync — Retroactive Record-Keeping

Use `task_sync` when work happened before Taskr was activated, or to retroactively close gaps in the task history. Three steps:

**Step 1 — Survey** (no args): lightweight summary of all task lists
```
task_sync(ruleContext="")
→ data.sync_check.task_lists[]: {id, name, status, task_count, done_count, open_count, wip_count}
```

**Step 2 — Drill in** (pass `task_list_id`): flat list of all tasks with title, description, status, level
```
task_sync(task_list_id="TL_...", ruleContext="")
→ data.sync_drill_down.tasks[]: {id, task_number, title, description, status, level, parent_id}
```

**Step 3 — Create done tasks** (pass `items[]`): creates tasks pre-marked as `done`
```
task_sync(items=[
  {
    "action": "create_done",
    "title": "...",
    "parentId": "TS_or_TL_...",
    "description": "...",
    "type": "implementation"
  }
], ruleContext="")
```

Each item requires `action: "create_done"` and `title`. `parentId` (TS_ or TL_) is required unless you provide `taskListTitle` to create a new task list — items without `parentId` default to that new list. Up to 50 items per call.

After syncing, attach CONTEXT or FINDING notes to the created tasks to capture the why behind the work.

---

## Cross-Session & Cross-Agent Continuity

Taskr state lives in the cloud — not in your context window. To resume work:

1. `search_notes` — find prior decisions, context, findings
2. `get_task_hierarchy(scope="project", depth=0)` — identify active task lists
3. `get_task(task_list_id="TL_...")` — pick up ownership where work stopped
4. Review CONTEXT and PROGRESS notes before executing

Any agent — different session, different machine, different model — can resume exactly where work stopped. The task list is the plan. Notes are the memory. Status is the state.

---

## Setup

When credentials are missing:

1. **Get credentials from user:**
   - Project ID: Projects page at https://taskr.one (format: `PR_00000000...`)
   - API Key: User avatar → API Keys menu (click eye icon or copy button)

2. **Configure via `gateway.config.patch`:**
   ```json
   {
     "skills": {
       "entries": {
         "taskr": {
           "env": {
             "MCP_API_URL": "https://taskr.one/api/mcp",
             "MCP_PROJECT_ID": "<project-id>",
             "MCP_USER_API_KEY": "<api-key>"
           }
         }
       }
     }
   }
   ```

3. **Verify:** Call `tools/list` and confirm `create_task` is present.

Users can create multiple projects for different work contexts.

**For mcporter/other MCP clients:**
```bash
mcporter config add taskr "$MCP_API_URL" \
  --header "x-project-id=$MCP_PROJECT_ID" \
  --header "x-user-api-key=$MCP_USER_API_KEY"
```

---

## Quick Reference

| Need | Tool |
|------|------|
| Create plan | `create_task` (taskListTitle + tasks[]) |
| Add context to new task list | `create_note` (type=CONTEXT, taskId=TL_) |
| Get next task | `get_task` |
| Mark done | `update_task` (taskId, status="done") |
| Mark skipped | `update_task` (status="skipped") + `create_note` (FINDING) |
| Browse task structure | `get_task_hierarchy` |
| Add subtask mid-execution | `create_task` (parentId=TS_or_TL_) |
| Document a finding | `create_note` (type=FINDING, taskId=TS_) |
| Phase milestone note | `create_note` (type=PROGRESS, taskId=TS_ phase task) |
| Retroactive history | `task_sync` (3-step) |
| Resume from prior session | `search_notes` → `get_task_hierarchy` → `get_task` |
| Find prior context | `search_notes` or `list_notes` |

**ruleContext values:** Pass the Rule ID from the schema (e.g. `RU-CTX-001`, `RU-PROC-001`, `RU-NOTE-001`). Pass `""` when the schema says "Leave this parameter blank."

AIエージェントの人気スキル

self-improving agentAIエージェント

開発中の学び、エラー、訂正を Markdown ファイルに記録し、再利用やワークスペースの記憶への昇格に備えます。

↓ 160.3万★ 4544pskoett
技能发现·Find SkillsAIエージェント

やりたいことを自然文で伝えると、ローカルとリモートのカタログからスキルを探して推薦するスキルです。

↓ 157.4万★ 1810桂皮
Agent BrowserAIエージェント

AI エージェント向けのブラウザ自動化コマンドラインスキルです。

↓ 122.3万★ 1001Nicolas Kleiber
Self-Improving + Proactive AgentAIエージェント

利用者の訂正と自己点検を層構造の記憶に蓄え、使われ方に応じて昇格や保管を行うスキルです。

↓ 69.6万★ 1622Iván
Proactive AgentAIエージェント

AI エージェントの先回り、作業記憶、振り返りの手順を計画するためのワークスペース設定ガイドです。

↓ 57.9万★ 916halthelobster
Skill CreatorAIエージェント

効果的な Claude スキルの作成・更新ガイドです。

↓ 47.7万★ 485chindden
Skill VetterAIエージェント

第三者のスキルを入れる前に、出所と権限と危険な作りを手順に沿って調べるスキルです。

↓ 44.5万★ 1328spclaudehome
唐朝诗仙李白 · Ai Agent SkillAIエージェント

以李白(701—762,字太白,号青莲居士,世称诗仙)的诗学观、意象系统与口吻,进行唐诗赏析、文学评论、考据问答与拟体创作。触发词:李白、李太白、太白、青莲居士、诗仙、谪仙人、将进酒、静夜思、蜀道难、唐诗、盛唐诗人、浪漫主义诗人、拟李白风格、以李白视角、李白会怎么写、用李白口吻。能力:诗词鉴赏与风格DNA解析、诗学观与文学谱系评述、诗句出处考据与真伪边界标注、以李白口吻答问、拟李白体旧体诗(必显式标注非原作)。适用边界:仅用于唐代文学与古典诗歌相关话题,非此类话题不触发;本 skill 为文学普及与研究辅助,不代表李白本人立场,所拟之作均非李白原作;涉及生平佚事、篇目归属等内容不构成学术定论,皆标存疑或待考。

↓ 43.6万★ 4user_ff4d9420

データ月:2026-10 · ダウンロード数とスター数は公開スキルレジストリ(ClawHub、SkillHub)から月単位で集計しています。詳しくはデータの方法論をご覧ください。