taskr（taskr 搜索）

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.

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.

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.

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:

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)

1. Configure via gateway.config.patch:

   {
     "skills": {
       "entries": {
         "taskr": {
           "env": {
             "MCP_API_URL": "https://taskr.one/api/mcp",
             "MCP_PROJECT_ID": "<project-id>",
             "MCP_USER_API_KEY": "<api-key>"
           }
         }
       }
     }
   }

1. Verify: Call tools/list and confirm create_task is present.

Users can create multiple projects for different work contexts.

For mcporter/other MCP clients:

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

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."