agentflow
Skill2 starsChanged 9 months agoMITView on GitHub
掌天瓶:异构多代理编排系统。当用户需要多代理协作时自动触发: - 说"启用掌天瓶"、"agentflow"、"多代理"、"并行处理" - 使用流水线指令("按 review 流水线..."、"按 implement 流水线...") - 说"@plan"触发 Autopilot 模式 - 说"并行修复"、"swarm"触发 Swarm 模式 关键词:掌天瓶、agentflow、流水线、pipeline、swarm、autopilot、并行、多代理
5 other projects use a version of this.
Adopted by
Most recently taken up by
- berabuddies/agentflowtheir own version, first seen Oct 8, 2026
- shouc/agentflowtheir own version, first seen Oct 8, 2026
- ConnorGriffin/follow-throughtheir own version, first seen Aug 4, 2026
- Mothilal-M/agentflow-skillstheir own version, first seen Apr 25, 2026
- lschGSL/claude_code_templatetheir own version, first seen Apr 9, 2026
Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- berabuddies/agentflow★ 1.4ksame name
- shouc/agentflow★ 1.4ksame name
- lschGSL/claude_code_template★ 1same name
- ConnorGriffin/follow-through★ 0same name
- Mothilal-M/agentflow-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository3 other files it gives its agents.
- .claude/agentflow/agents.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- switch-agentflowSkill
The file
Files
掌天瓶 (Agentflow) — 异构多代理系统
代理委派策略
当需要执行以下类型的任务时,使用对应的子代理:
| 任务类型 | 委派给 | 模型路由 | 权限 |
| 后端代码实现 | backend-coder | Codex (via CCR) | Server-side only |
| 前端 UI/UX 实现 | frontend-coder | Gemini Pro (via CCR) | Client-side only |
| 代码审核/Review | reviewer | Codex (via CCR) | Read-only + test |
| 技术调研 | researcher | Gemini Flash (via CCR) | Read-only + web |
| 代码搜索/文件查找 | explorer | Gemini Flash (via CCR) | Read-only |
| 文档撰写 | doc-writer | Gemini Flash (via CCR) | .md only |
| 需求规划 | planner | Claude Opus (native) | Read-only |
| 核心代码 | 主代理自行处理 | Claude Opus (native) | Full |
完整注册表见 .claude/agentflow/agents.md。
三种执行模式
Pipeline(流水线)
多阶段链式执行,每阶段委派给不同代理。
触发方式:
- "按 review 流水线审查 {target}"
- "按 implement 流水线实现 {feature}"
- "按 research 流水线调研 {topic}"
- "按 debug 流水线修复 {bug}"
可用预设(定义在 .claude/agentflow/workflows/):
| 预设 | 阶段 |
| review | explore → review → fix → verify |
| implement | plan → implement → review |
| research | explore → research → summarize |
| debug | explore → analyze → fix |
执行流程:
- 读取
.claude/agentflow/workflows/{预设}.yaml - 设置 workflow-state.json:
active=true, 记录 stages - 按顺序委派每个 stage 给对应代理
- 每个 stage 完成后更新 state
- Stop Hook 自动阻止中途停止
- 全部完成 →
active=false→ 允许停止
Autopilot(自主模式)
planner 规划 → 自动按计划执行。
触发方式:
- "@plan {需求描述}"
- "规划并实现 {feature}"
执行流程:
- 调用 planner 子代理生成
current-plan.md(含 TODO 列表) - 按 TODO 逐项委派给合适的代理
- 每完成一项更新 checkbox
- Stop Hook 检查未勾选 TODO → 阻止停止
- 全部完成 → 允许停止
Swarm(蜂群并行)
多 worker 从共享任务池并行处理。
触发方式:
- "并行修复所有 {问题类型}"
- "并行处理以下任务: {task1}, {task2}, ..."
执行流程:
- 分解为原子任务,写入 task-pool.json
- 启动多个子代理(background task)
- 每个子代理调用
claim-task.sh原子认领 - 完成后调用
complete-task.sh - Stop Hook 检查 task-pool → 阻止停止
- 全部完成 → 允许停止
任务池脚本(.claude/agentflow/scripts/):
create-pool.sh <name> <tasks-file>— 创建任务池claim-task.sh <worker-id> [agent]— 原子认领complete-task.sh <task-id> [result]— 标记完成release-timeout.sh [seconds]— 释放超时任务pool-status.sh— 查看状态
自动继续规则
掌天瓶启用时,Stop Hook 在以下条件阻止主代理停止:
- 工作流 active + 有 pending stages
- 任务池有 pending/claimed 任务
- 计划文档有未勾选 TODO
紧急退出:touch /tmp/FORCE_STOP
结果输出规范
所有子代理输出写入 .claude/agentflow/results/:
- 代码任务:
{task-id}.json(含 verdict, issues, tests_passed) - 调研任务:
{task-id}.md(Markdown 报告) - 搜索任务:
{task-id}.json(文件列表和代码片段)
工作流状态管理
状态文件:.claude/agentflow/state/workflow-state.json
{
"active": true,
"mode": "pipeline",
"workflow": "review",
"current_stage": "explore",
"completed_stages": [],
"pending_stages": ["review", "fix", "verify"],
"target": "src/auth/"
}启动工作流时写入此文件,完成后设 active: false。
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

