什么是 任务恢复？

任务恢复（Task Resume）是一款先进的状态管理技能，旨在消除因上下文切换引起的摩擦。在任何复杂的 AI 智能体环境中，用户经常在当前任务完成前提出新请求。该技能通过将中断上下文持久化到工作区全局队列，确保智能体不会丢失其原始目标。通过将其集成到 Openclaw Skills 中，开发人员可以确保高保真地恢复任务状态，包括验收标准和完成工作所需的准确后续步骤。

该技能基于“默认入队”哲学运行，这意味着除非明确说明，否则它会将大多数主题切换视为潜在中断。这为高优先级任务提供了安全保障，防止它们在繁忙的开发阶段被边缘化或遗忘。

下载入口:https://github.com/openclaw/skills/tree/main/skills/richardsun700/task-resume

安装与下载

1. ClawHub CLI

从源直接安装技能的最快方式。

2. 手动安装

将技能文件夹复制到以下位置之一

优先级：工作区 > 本地 > 内置

3. 提示词安装

将此提示词复制到 OpenClaw 即可自动安装。

任务恢复 应用场景

• 当用户切换到新的无关请求时，自动保存进度。
• 在多智能体或多会话协作期间管理任务优先级。
• 防止在临时上下文切换或系统触发的中断期间丢失工作。
• 在完成高优先级热修复或查询后立即恢复后台任务。

1. 智能体识别具有未完成验收标准的活动任务。
2. 收到无关的用户消息后，智能体将任务上下文持久化到共享的全局队列文件中。
3. 智能体处理新的中断或优先级请求。
4. 活动任务完成后，智能体自动从队列中弹出最早的项目。
5. 智能体恢复之前的上下文并继续执行，并向用户宣告任务恢复。

任务恢复 配置指南

要将此功能集成到您的 Openclaw Skills 工作区，请确保配置如下：

1. 在 memory/task-resume-queue.json 初始化全局队列存储。
2. 部署用于处理队列操作的辅助脚本：

# 检查任务队列的当前状态
python3 skills/task-resume/scripts/task_resume_queue.py status

3. （可选）为看门狗自动继续功能配置每 30 分钟运行一次的 cron 任务。

任务恢复 数据架构与分类体系

该技能通过统一的 JSON 模式管理任务状态，允许主智能体和克隆智能体之间进行跨会话可见性。

name: task-resume
description: Automatic interrupted-task resume workflow with queueing and recovery. Use when a user asks to resume interrupted work after temporary context switches, protect priority tasks from drift, or enforce "finish current task then auto-resume previous interrupted task" behavior.

Task Resume

Use this skill to ensure interrupted tasks are recovered automatically.

Defaults (MANDATORY)

• Make interruption enqueue the default behavior.
• On any non-explicit context switch, auto-enqueue the active unfinished task.
• Enqueue at message-time (immediately when interruption is detected), not only on periodic checks.
• Use one shared queue file for all sessions/clones so the view is unified.

Rules

• Treat user-explicit commands (cancel, pause, change priority, do it tomorrow) as overrides.
• Treat all other topic switches as potential interruptions.
• Before switching topics, persist interruption context to queue.
• After completing the active task, immediately resume the oldest queued interrupted task.
• If queue is empty, do nothing.

Shared Queue Storage (cross-session)

• Canonical queue file: memory/task-resume-queue.json
• This file is workspace-global and shared across main + clone + group sessions.
• Always pass source with session/channel identity.
• Use helper script: scripts/task_resume_queue.py

Interruption Detection

Consider a task interrupted when all are true:

1. There is an active task with unfinished acceptance criteria.
2. A new user request is unrelated to finishing that active task.
3. User did not explicitly cancel/pause/defer the active task.

Message-Time Enforcement (required)

Before handling every new user message:

1. Check whether an active task exists and is unfinished.
2. If the incoming message is an unrelated request and no explicit override is present, enqueue immediately.
3. Only then switch to the new request.

This prevents queue misses caused by timing gaps.

Log-based Recovery (ENOENT-safe)

When recovery needs session .jsonl context, use:

python3 skills/task-resume/scripts/task_resume_queue.py recover r
  --log "~/.openclaw/agents/main/sessions/.jsonl" r
  --title "" r
  --acceptance "" r
  --source "" r
  --session ""

If the log file is missing (ENOENT), treat it as expected and continue (skipped_missing_log), do not raise alert-level failure.

On Interruption (auto-enqueue)

Run immediately at interruption detection:

python3 skills/task-resume/scripts/task_resume_queue.py add r
  --title "" r
  --context "" r
  --acceptance "" r
  --source "" r
  --session ""

Then acknowledge briefly: queued + will auto-resume.

On Active Task Completion (resume)

Run:

python3 skills/task-resume/scripts/task_resume_queue.py pop

• If one item is returned, resume it immediately and announce: Resuming previously interrupted task: