OpenClaw： benign
VirusTotal： benign
StaticScan： clean
OpenClaw： benign
This is an instruction-only collaboration governance skill whose requirements and instructions match its stated purpose and request no extra credentials or installs.
VirusTotal： benign VT 报告
静态扫描： clean
No suspicious patterns detected.
README 未提供
无文件信息
{
"latestVersion": {
"_creationTime": 1774206771885,
"_id": "k978mjjagf61kjbagay4gsnhks83crhr",
"changelog": "中文介绍：\n\nContext Flow\n多 agent 协作时的上下文管理技能。解决协作过程中最常见的四类问题：上下文膨胀、任务串线、群聊失控、中断后无法恢复。\n\n它能做什么\n给子agent 下发任务时，自动使用任务卡片格式（目标、范围、约束、不要做、输出格式），避免甩一整段聊天记录\n长任务按阶段推进，每阶段结束后先产出短版收口，再进入下一阶段\n检测上下文过载信号（单agent 承载过多子目标、subagent 开始推理全局任务线、后台插件超时导致连接断开）\n群聊变乱时提供 reset 流程，中断后提供基于checkpoint 的恢复机制\n提供反模式清单，避免常见协作陷阱\n包含哪些文件\nCopyCopied!\ncontext-flow\/\n├── SKILL.md # 技能入口，定义触发条件和引用关系\n├── PACKAGING.md # 打包指引（开发者用）\n└── references\/\n ├── templates.md # 任务卡片、收口模板、恢复模板、状态机规则、最低证据标准\n ├── anti-patterns.md # 反模式清单（上下文膨胀、任务串线等）\n ├── feishu-usage.md # 飞书群聊场景的具体用法\n └── lessons.md # 恢复和反思的实战经验\n怎么用\n安装后，agent 会在以下场景自动激活：\n\n你说\"给planner 派个任务\" → agent 用任务卡片格式下发，而不是复制粘贴聊天记录\n你说\"这个任务太长了，先收口\" → agent 按分阶段收口模板输出当前进度\n群聊里消息越来越乱 → agent 检测到过载信号，建议做一次 checkpoint或 reset\n上次对话中断了，你说\"继续\" → agent 从 checkpoint 恢复，而不是重新推理整个任务\n也可以直接对 agent 说：\n\n帮我给子 agent 写一个任务卡片\n当前任务做个 checkpoint\n群聊上下文太乱了，帮我整理一下\n上次中断了，帮我恢复\n安装\nbashCopyCopied!\nnpx clawhub@latest install context-flow\n或手动复制到 ~\/.openclaw\/skills\/。无额外依赖。\n\n配合使用\n可独立使用。如果同时安装了 collab-setup，两者互补：collab-setup 管\"怎么配置和修复协作基础设施\"，context-flow 管\"协作过程中怎么管理上下文\"。\n\nEnglish description：\n\nContext Flow\nContext management skill for multi-agent collaboration. Solves the four most common collaboration problems: context bloat, task line crossing, chaotic group chats, and inability to recover after interruptions.\n\nWhat it does\nWhen dispatching tasks to child agents, automatically uses task card format (goal, scope, constraints, don'ts, output format) instead of dumping raw chat history\nBreaks long tasks into stages with short closure summaries between each stage\nDetects context overload signals (single agent carrying too many sub-goals, subagent reasoning about the full task line, background plugin timeouts causing disconnection)\nProvides reset flow when group chats go chaotic, and checkpoint-based recovery after interruptions\nIncludes anti-pattern checklist to avoid common collaboration pitfalls\nFile structure\nCopyCopied!\ncontext-flow\/\n├── SKILL.md # Skill entry point with trigger conditions and references\n├── PACKAGING.md # Packaging guide (for developers)\n└── references\/\n ├── templates.md # Task cards, closure templates, recovery templates, state machine rules, minimum evidence rules\n ├── anti-patterns.md # Anti-pattern checklist (context bloat, task crossing, etc.)\n ├── feishu-usage.md # Feishu group chat specific usage\n └── lessons.md # Recovery and reflection lessons from real-world experience\nHow to use\nAfter installation, the agent activates automatically in these scenarios:\n\nYou say \"dispatch a task to planner\" → agent uses task card format instead of copy-pasting chat history\nYou say \"this task is too long, let's wrap up\" → agent outputs current progress using staged closure template\nGroup chat gets messy → agent detects overload signals and suggests a checkpoint or reset\nPrevious conversation was interrupted, you say \"continue\" → agent recovers from checkpoint instead of re-reasoning the entire task\nYou can also directly ask:\n\nWrite a task card for the sub-agent\nCreate a checkpoint for the current task\nGroup chat context is too messy, help me organize\nLast session was interrupted, help me recover\nInstall\nbashCopyCopied!\nnpx clawhub@latest install context-flow\nOr manually copy to ~\/.openclaw\/skills\/. No additional dependencies.\n\nCompanion skill\nWorks standalone. If collab-setup is also installed, they complement each other: collab-setup handles \"how to configure and fix collaboration infrastructure\", context-flow handles \"how to manage context during collaboration\".",
"changelogSource": "user",
"createdAt": 1774206771885,
"version": "1.0.2"
},
"owner": {
"_creationTime": 0,
"_id": "publishers:missing",
"displayName": "OldBulls",
"handle": "oldbulls",
"image": "https:\/\/avatars.githubusercontent.com\/u\/9461547?v=4",
"kind": "user",
"linkedUserId": "kn752wyf3z2v2gfe0dm9cewjas839tk2"
},
"ownerHandle": "oldbulls",
"skill": {
"_creationTime": 1774051717542,
"_id": "kd766xa8br9bqt6xgr2nex50j183a11h",
"badges": [],
"createdAt": 1774051717542,
"displayName": "Context Flow",
"latestVersionId": "k978mjjagf61kjbagay4gsnhks83crhr",
"ownerUserId": "kn752wyf3z2v2gfe0dm9cewjas839tk2",
"slug": "context-flow",
"stats": {
"comments": 0,
"downloads": 58,
"installsAllTime": 0,
"installsCurrent": 0,
"stars": 0,
"versions": 3
},
"summary": "Manage context flow for group collaboration, child agents, and subagents. Use when coordinating multi-agent work, handing off tasks, preventing context overl...",
"tags": {
"chat-ops": "k9792a7ej3x9721ynzzfwfj12d83ad7b",
"checkpoint": "k978mjjagf61kjbagay4gsnhks83crhr",
"collaboration": "k978mjjagf61kjbagay4gsnhks83crhr",
"context-management": "k978mjjagf61kjbagay4gsnhks83crhr",
"governance": "k9792a7ej3x9721ynzzfwfj12d83ad7b",
"group-chat": "k978mjjagf61kjbagay4gsnhks83crhr",
"handoff": "k97e1dann7xq3jmy7ke202zh1183a43n",
"latest": "k978mjjagf61kjbagay4gsnhks83crhr",
"multi-agent": "k978mjjagf61kjbagay4gsnhks83crhr",
"overload-prevention": "k978mjjagf61kjbagay4gsnhks83crhr",
"recovery": "k978mjjagf61kjbagay4gsnhks83crhr",
"resume": "k9792a7ej3x9721ynzzfwfj12d83ad7b",
"task-cards": "k978mjjagf61kjbagay4gsnhks83crhr",
"task-handoff": "k9792a7ej3x9721ynzzfwfj12d83ad7b",
"templates": "k97e1dann7xq3jmy7ke202zh1183a43n",
"workflow": "k97e1dann7xq3jmy7ke202zh1183a43n"
},
"updatedAt": 1774207907930
}
}