arc-wake-state
由 trypto1019 提供
Notes & PKM简介
Persist agent state across crashes, context deaths, and restarts.
描述
该技能的功能
Saves agent state to structured files so autonomous agents can resume work after context window resets or crashes. Provides a persistent task queue, named checkpoints, and crash detection based on heartbeat timestamps.
Structured state files and crash detection give agents a reliable handoff without requiring external databases or custom persistence code.
何时使用
- Resuming a multi-hour agent task after context window resets
- Detecting whether a previous session crashed or completed cleanly
- Creating a named checkpoint before running a risky database migration
- Maintaining a task queue that survives repeated agent restarts
- Rolling back agent state to a known-good checkpoint after a failed operation
示例工作流
User asks: Resume the task you were working on before the context reset
- 1
Run crash-check to determine if the last session ended cleanly or crashed
- 2
Run read to retrieve current status, notes, and key-value pairs
- 3
Run tasks to list all pending items from the persistent queue
- 4
Identify the last recorded status and continue from that point
- 5
Call heartbeat on each subsequent loop iteration to enable future crash detection
Agent identifies prior status as 'Building budget tracker skill', picks up pending tasks from the queue, and continues without manual re-briefing
安装
clawhub install trypto1019/arc-wake-state运行 CLI 命令,或把这段提示词粘贴到 Claude Code、Cursor、Codex 或你在用的任意编程智能体里。