任务打断机制

管理和中断长时间运行任务，支持用户命令终止、资源清理和任务状态保存，防止任务卡死或无限执行。

This skill appears coherent and implements a simple flag-file + polling interrupt pattern. Before installing, verify: (1) how sessionId is mapped to processId in your environment — ensure only the intended child process can be signalled; (2) file permissions and /tmp isolation — flag files created in /tmp are world-visible by default, consider using a per-agent temp directory or restrictive permissions; (3) that checkpoint/save logic is implemented in your long-running tasks so interruptions are safe; (4) scripts are placed where the agent expects and marked executable; (5) test in a staging environment to confirm no accidental PID collisions or privilege escalations. If you need stronger guarantees, consider authenticated/IPC-based signaling instead of plain /tmp files and process.kill.

Type: OpenClaw Skill Name: task-interrupt Version: 1.0.1 The 'task-interrupt' skill provides a legitimate mechanism for managing and terminating long-running Agent tasks. It uses flag files in /tmp (e.g., /tmp/agent-stop-{sessionId}.flag) and shell scripts (create-stop-flag.sh, check-stop-flag.sh) to signal interruptions, save checkpoints, and perform resource cleanup. The logic is transparently documented in SKILL.md and instructions.md, and the file operations are restricted to task-specific temporary files, showing no signs of malicious intent or data exfiltration.