subagent-orchestrator (汪哈哈版)

将复杂大型任务拆解为多个子任务，通过 Subagent 并行/串行执行，三文件持久化防丢失，Main Agent 统筹整合。适用于任何需要多步骤、多信息源、长时间运行的任务，也适用于即将超出上下文窗口时的任务分流。触发词：任务分工、子任务编排、多agent协作、上下文溢出、任务分流、context快满了。

This skill is coherent with its purpose but will perform persistent and powerful actions at runtime: it will create and read files under ~/.openclaw/workspace/tasks, run simple shell checks (ls/mkdir), spawn subagent sessions (including passing full task context), and may schedule periodic cleanup via cron. Before installing or enabling: (1) Confirm you trust the agent to create/modify files in your home directory and to execute basic shell commands. (2) Review or sandbox the workspace path (~/.openclaw/workspace/tasks) so only expected files are written. (3) Monitor any cron entries the agent may create and remove them if undesired. (4) Ensure no sensitive credentials are stored in referenced files or in task parameters passed to subagents. (5) If you want to restrict persistence, require explicit user confirmation before the skill writes files or schedules cron jobs (the SKILL.md already requires confirmation for nested spawning).

Type: OpenClaw Skill Name: subagent-orchestrator-whhh Version: 1.3.0 The skill is a complex task orchestrator that uses subagents and a three-file persistence system to manage long-running tasks and mitigate context window limits. It is classified as suspicious due to high-risk capabilities, specifically the inclusion of a 'Daily Session Cleanup' feature in SKILL.md that uses the `cron` tool to schedule the execution of a shell script (~/scripts/scan_sessions.sh) and perform automated file deletions. While these actions are aligned with the stated purpose of system maintenance and require user confirmation, the combination of persistence via cron, shell execution, and automated data removal represents a significant attack surface without clear safeguards for the external script's content.