agentic-workflow（Agent 工作流程）

Agentic Workflow (Kanban & Heartbeat System)

This skill enables an OpenClaw agent to operate continuously in the background using a State Machine (Kanban board) driven by heartbeats, and ensures high-quality output through a Maker-Checker (QA subagent) verification loop.

Core Components

1. The Task Board (TASK_BOARD.yaml): The single source of truth for all tasks.
2. The Heartbeat (HEARTBEAT.md): The cron-engine that reads the board and executes tasks without user intervention.
3. The Checker (QA Subagent): The sessions_spawn mechanism used to verify results before showing them to the user.

Implementation Steps (How to install this for an agent)

When a user asks you to "set up the task system" or "agentic workflow", follow these steps in their workspace:

1. Create TASK_BOARD.yaml

Create this file in the workspace root:

# Master Task Board
# Status Enum: TODO, IN_PROGRESS, QA_REVIEW, DONE, BLOCKED

current_sprint:
  active: false
  focus: "General"

tasks:
  - id: T-001
    title: "Example Task"
    status: TODO
    created_at: "YYYY-MM-DD"
    description: "What needs to be done."
    history: []

2. Update HEARTBEAT.md

Ensure the agent's HEARTBEAT.md contains the following Executor instruction (usually at the top or highest priority):

### Task Board Executor (Highest Priority)
**Trigger**: Every heartbeat
**Action**:
1. Read `TASK_BOARD.yaml`.
2. If a task is `IN_PROGRESS`, continue its next step and update the `history` in YAML.
3. If no `IN_PROGRESS`, pick the highest priority `TODO` task, set to `IN_PROGRESS`, and begin.
4. When a task step yields a deliverable, set status to `QA_REVIEW`. Use `sessions_spawn(runtime="subagent")` to spawn a strict QA Checker agent. Give it the original goal and the output.
5. If the QA Checker approves, set status to `DONE` and notify the user. If it fails, fix the issue. If it fails 3 times, set to `BLOCKED` and notify the user.
6. If everything is running smoothly or waiting, DO NOT message the user. Reply `HEARTBEAT_OK` to stay silent.

The Maker-Checker Loop (Crucial!)

When you (the Maker) finish a piece of work (e.g., generating a PDF, writing a script), you must not immediately tell the user. Instead, you must spawn a subagent to act as the Checker.

Example sessions_spawn payload for the Checker:

{
  "task": "You are a strict QA inspector. Review this output: [Output]. Does it perfectly meet these requirements: [Requirements]? Reply ONLY with 'PASS' or a list of specific flaws to fix.",
  "runtime": "subagent",
  "mode": "run",
  "agentId": "distiller"  // Or the default subagent
}

Golden Rules for the Agent

• Silence is Golden: Never message the user just to say "I am working on step 2." Only message them when a task hits DONE or BLOCKED.
• Read Before Acting: Always read TASK_BOARD.yaml upon waking up (heartbeat) to know your current state.
• Self-Correction: Let the QA subagent hurt your feelings. Fix the code/output internally before bothering the human.