Session Resume

会话恢复。Gateway 重启或 session 中断后，自动恢复任务上下文并向用户汇报进度。通过 .task-state.json 持久化任务状态，确保长时间任务不会因断线而丢失。触发词：恢复任务、resume、任务状态、断线恢复、session resume。也可在每次 session 启动时自动检测。

Security Scan

The skill's requirements and instructions are consistent with its stated purpose of persisting and restoring long-running task state; no unrelated credentials, installers, or network endpoints are requested.

Purpose & Capability

Name, description, and runtime instructions all focus on saving/restoring a .task-state.json file in the OpenClaw workspace to resume long-running tasks. There are no unrelated env vars, binaries, or install actions demanded — the declared capabilities match what is needed for this functionality.

Instruction Scope

SKILL.md explicitly instructs the agent to read/write ~/.openclaw/workspace-main/.task-state.json and to be invoked at session start (recommended insertion into AGENTS.md). This is in-scope for a resume feature, but the instructions do not mention any protections for sensitive data stored in the state file (encryption, redaction, or access controls), which means sensitive context could be persisted unintentionally.

Install Mechanism

There is no install spec embedded in the skill (instruction-only). README suggests installing from ClawHub or cloning a GitHub repo; those are expected and do not embed arbitrary download/execution logic in the skill itself.

Credentials

The skill requests no credentials or special environment variables, which is appropriate. However, it persists 'context' fields that may contain file paths, config values, or intermediate results; storing such data locally can expose secrets if users or other processes can access the workspace file. The lack of guidance about redaction or file permissions is a proportionality concern to consider.

Persistence & Privilege

The skill does not request elevated privileges, does not set always:true, and only recommends being invoked at session start. Its suggested behavior — reading/writing a state file in the user's OpenClaw workspace — is a normal level of persistence for this feature.

Assessment

This skill appears to do what it says (save/restore a .task-state.json to resume long-running tasks). Before installing: 1) Verify the source (check the GitHub repo / ClawHub package owner) since README points to an external repo. 2) Inspect any example or existing .task-state.json files to see what 'context' they store; avoid saving secrets (API keys, passwords, private keys) into task state. 3) Limit file access (store the file under ~/.openclaw with restrictive permissions, e.g., chmod 600) or consider adding encryption/redaction if your tasks include sensitive data. 4) If you enable automatic detection at session start (editing AGENTS.md), be aware the agent will read the state file automatically — ensure only trusted agents run on that host. 5) If you need stronger guarantees, request that the skill include explicit redaction/encryption steps or only record non-sensitive metadata. These precautions will reduce the chance of accidentally persisting sensitive information.

Session Resume — 会话恢复

Gateway 重启、网络断线、session 压缩后，自动恢复任务上下文。

什么时候用

• 每次新 session 启动时（自动检测）
• 用户说"恢复任务"、"resume"、"断线恢复"
• Gateway 重启后的第一次对话
• 长时间任务中途被中断

核心理念

AI Agent 执行长时间任务时最怕断线——Gateway 重启、session 被压缩、网络中断， 任务进度全部丢失，用户不得不重新描述需求。

Session Resume 通过任务状态持久化解决这个问题：

• 执行任务时定期保存进度到文件
• 新 session 启动时自动读取并恢复
• 向用户汇报：之前在做什么、做到哪了、还剩什么

执行流程

Phase 1 — 检测（Detect）

新 session 启动时检查：

1. 读取 ~/.openclaw/workspace-main/.task-state.json
2. 如果文件存在且内容有效，说明有未完成的任务
3. 如果文件不存在或为空，说明没有待恢复的任务（跳过后续步骤）

Phase 2 — 恢复（Restore）

解析任务状态文件，提取：

• 任务描述：用户原始请求是什么
• 当前进度：已完成的步骤
• 未完成项：还剩哪些步骤
• 关键上下文：文件路径、配置值、中间结果等
• 中断原因：如果能判断的话（压缩/重启/超时）

Phase 3 — 汇报（Report）

向用户发送恢复报告：

🔄 任务恢复
━━━━━━━━━

📋 之前的任务：
部署新版本到生产环境

✅ 已完成：
- 代码拉取和构建
- 单元测试通过
- Docker 镜像构建完成

⏳ 待完成：
- 推送镜像到仓库
- 更新 K8s 配置
- 执行滚动更新

💡 是否继续执行？

Phase 4 — 继续或放弃（Resume / Abandon）

• 用户说"继续" → 从断点处继续执行
• 用户说"放弃"或"不用了" → 清理状态文件，开始新任务
• 用户不回应 → 保持状态文件不动，下次再问

任务状态文件格式

文件路径：~/.openclaw/workspace-main/.task-state.json

{
  "version": 1,
  "taskId": "uuid-here",
  "createdAt": "2026-04-02T10:30:00+08:00",
  "updatedAt": "2026-04-02T11:15:00+08:00",
  "description": "用户的原始任务描述",
  "status": "in_progress",
  "steps": [
    {"name": "步骤1", "status": "done", "result": "简要结果"},
    {"name": "步骤2", "status": "done", "result": "简要结果"},
    {"name": "步骤3", "status": "pending"},
    {"name": "步骤4", "status": "pending"}
  ],
  "context": {
    "key1": "value1",
    "key2": "value2"
  },
  "lastCheckpoint": "步骤2完成后"
}

使用规范

保存任务状态（执行长时间任务时）

在执行多步骤任务时，每完成一个关键步骤后更新状态文件：

1. 任务开始时创建 .task-state.json
2. 每完成一个步骤，更新对应 step 的 status 和 result
3. 任务全部完成后删除状态文件

自动检测（session 启动时）

建议在 AGENTS.md 的启动序列中加入：

5. 检查 .task-state.json 是否存在，如有则执行 session-resume skill 恢复任务

规则

• 只读取，不猜测：只恢复文件中明确记录的信息，不猜测未记录的进度
• 先汇报，后执行：恢复后先告知用户，等用户确认再继续
• 及时更新：每个关键步骤完成后都要更新状态文件
• 任务完成即清理：全部完成后删除状态文件，避免下次误恢复
• 不影响正常流程：没有状态文件时，完全不影响正常的 session 启动

与其他 Skill 的配合

• Smart Compact：压缩前先保存任务状态，压缩后自动恢复
• Memory-Dream：任务状态是临时的（完成即删），长期信息由 Dream 整合到 MEMORY.md