clawclau

What This Skill Does

ClawClau (v2.0.0) is a robust, asynchronous task orchestration system for OpenClaw that enables users to delegate complex Claude Code tasks to the background. By leveraging tmux for session management, it allows for long-running workflows that persist independently of the main agent process. It features a sophisticated state-machine approach, providing real-time progress reports, automated completion notifications, and granular control over background processes. Whether you need to run a multi-hour data analysis, deep code research, or batch file processing, ClawClau ensures these tasks remain isolated, monitorable, and interruptible.

Installation

Installation requires standard CLI utilities available on most Unix environments. First, ensure tmux and jq are installed via brew install tmux jq. Next, prepare the environment by executing mkdir -p ~/.clawclau/logs ~/.clawclau/prompts. You may optionally configure notifications by appending notify_chat = oc_xxxxxxxx to ~/.clawclau/config. Finally, verify that your $CC_SCRIPTS environment variable points to your clawclau/scripts directory, allowing OpenClaw to call the underlying shell scripts seamlessly.

Use Cases

ClawClau is designed for high-latency or compute-heavy tasks that would otherwise hang the primary agent thread. Use it when:

• Executing large-scale codebase refactoring or automated debugging.
• Performing multi-step research reports that require significant context processing.
• Running batch tasks that require periodic status polling without holding an active session open.
• Requiring "steerable" tasks where you need to guide the agent mid-process based on its output.

Example Prompts

1. "让乌萨奇去分析一下 src/ 目录下所有的组件，整理一份详细的性能瓶颈报告，任务名设为 perf-audit。"
2. "查看一下后台名为 research-task 的任务进度，如果已经完成，把结果发送给我。"
3. "终止 heavy-import 任务，这个任务跑得太慢了，我们需要手动介入纠偏。"

Tips & Limitations

• Persistence: Always use unique Task IDs to avoid collisions in the active-tasks.json registry.
• Steerability: If you anticipate needing to pivot the agent's logic mid-task, always spawn with the --steerable flag; non-steerable tasks do not support claude-steer.sh.
• Monitoring: Set up the claude-monitor.sh script via a cron job to ensure that failed or hanging sessions are properly garbage-collected and notified to the agent.
• Safety: Because this tool interacts with the filesystem via Claude Code, ensure you have appropriate backups when running tasks in critical directories.