This skill provides a complete multi-agent collaboration system for product-driven development loops

Name: This skill provides a complete multi-agent collaboration system for product-driven development loops
Author: weitanai

作者 Guangsheng Ye · GitHub ↗ · v1.0.0 · MIT-0

cross-platform ⚠ suspicious

405

总下载

当前安装

版本数

在 OpenClaw 中安装

/install devloop-agent-pack

功能描述

Complete multi-agent collaboration workflow for product-driven development loops. Covers the full lifecycle from product discovery through architecture desig...

使用说明 (SKILL.md)

DevLoop Workflow — 多 Agent 协作系统

产品驱动开发闭环 — 开箱即用的多 Agent 协作系统，覆盖从产品热点探索到开发测试上线的完整生命周期。

Agent 清单

Agent	ID	Emoji	职责
Product	`devloop-product`	🎯	每日 AI 热点探索、产品方向讨论、PRD 生成
Core Dev	`devloop-core-dev`	🧠	架构设计（7 维度）、设计文档、Dev 调度
Dev	`devloop-dev`	⚡	精准编码、多实例并行、按设计文档实现
Test	`devloop-test`	🧪	测试先行、Bug 趋势追踪、代码审查
Marketing	`devloop-marketing`	📣	商业化调研、宣传策略、文案制作
Research	`devloop-research`	🔬	深度调研、竞品分析、技术评估

核心工作流

阶段一：产品发现
  [Cron 触发] → 🎯 Product（热点探索 + 历史对比）

阶段二：方向讨论与 PRD
  用户 ↔ 🎯 Product 多轮讨论 → 确认方向 → 生成 PRD
       → 通知 🧠 Core Dev + 📣 Marketing

阶段三：设计与调度
  🧠 Core Dev（7维度讨论→设计文档→复杂度评估→调度）
       → 分配任务给 ⚡ Dev (×N)
       → 通知 🧪 Test 准备测试规格

阶段四：测试先行
  🧪 Test（读取设计文档 → 生成测试规格 → 通知 Dev 参考）

阶段五：编码实现
  ⚡ Dev ×N（严格按设计文档编码 → Conventional Commits → 报告完成）

阶段六：测试与合并
  🧪 Test（代码审查 + 测试执行 + Bug 记录）
  🧠 Core Dev 确认质量 → 合并到 main
       → 📣 Marketing（上线宣传）

完整的阶段细节、消息路由表和并行冲突预防规则，参见 references/collaboration-protocol.md。

通用工作规范

Session 启动

读取 SOUL.override.md（如存在，替代 SOUL.md），否则读取 SOUL.md
读取 USER.md
读取 memory/YYYY-MM-DD.md（今天 + 昨天）
仅主 session：读取 MEMORY.md

首次启动：若 BOOTSTRAP.md 存在，按其指引初始化后删除。

记忆管理

类型	路径	说明
每日笔记	`memory/YYYY-MM-DD.md`	当日工作日志（平面文件，不建子目录）
主题笔记	`memory/YYYY-MM-DD-\x3Csuffix>.md`	按主题的专项日志
长期记忆	`MEMORY.md`	跨 session 持久知识（仅主 session 加载，含私人上下文）

想记住的东西必须写文件。"心里记住"在 session 结束后消失。

Agent 协作

通过 sessions_send 通信。共享文件通过各自 workspace 的 shared/ 目录（只读消费，不修改，不用 ../ 路径）。

自定义

优先级：SOUL.override.md > SOUL.md > 各 Agent .md

安全

不泄露私人数据
破坏性命令先确认
trash > rm

模板文件

所有模板位于本 skill 的 assets/templates/ 目录。Agent 首次创建文件时，读取对应模板并按实际内容填充。

模板文件	用途	使用者
`project-structure.template.md`	项目知识库结构	Product, Marketing
`design-doc.template.md`	功能设计文档	Core Dev
`design-index.template.md`	设计文档索引	Core Dev
`test-spec.template.md`	测试规格文档	Test
`daily-report.template.md`	每日测试报告	Test
`bug-tracker.template.md`	Bug 追踪数据库	Test
`review-notes.template.md`	PR 审查笔记	Test
`bug-trend.template.md`	Bug 趋势汇总（MEMORY.md 用）	Test
`memory-tracking.template.md`	调研方向追踪表（MEMORY.md 用）	Product, Marketing

参考资料

需要深入了解时，加载以下 references 文件：

文件	何时加载	内容
`references/collaboration-protocol.md`	需要了解 Agent 间通信细节、消息格式、文件共享规则、完整工作流阶段细节时	完整协作协议、消息路由表、共享目录约定、并行冲突预防
`references/agent-design-background.md`	需要理解设计决策背景、处理边缘情况时	各 Agent 的设计理念、权限说明、边缘情况处理方案

快速查找关键信息：

Agent 间消息格式：搜索 sessions_send in references/collaboration-protocol.md
文件共享规则：搜索 shared/ in references/collaboration-protocol.md
并行冲突处理：搜索 交叉 in references/collaboration-protocol.md
各 Agent 权限设计：搜索 权限设计 in references/agent-design-background.md
边缘情况处理：搜索 边缘情况 in references/agent-design-background.md

前置要求

OpenClaw 2026.3.x+
Linux / macOS（Windows 需 WSL）

安全使用建议

This skill appears coherent with its stated purpose, but it grants development privileges that can modify your repository and run shell commands. Before installing: (1) Review any BOOTSTRAP.md, SOUL.override.md, MEMORY.md and USER.md files in the workspace — they will be loaded and can change agent behavior (BOOTSTRAP may be executed and deleted). (2) If you do not want agents to run shell/git or modify main, restrict/remove Bash/exec permissions for Core Dev/Dev/Test or run the skill in an isolated sandbox repository. (3) Accept network access for Product/Research/Marketing agents (they perform web_search/web_fetch) only if that’s expected. (4) Verify the upstream repository/homepage and trust the author; inspect any bootstrap or override scripts before allowing them to run. If you want lower risk, use the documentation/templates without enabling the agents' exec permissions.

功能分析

Type: OpenClaw Skill Name: devloop-agent-pack Version: 1.0.0 The devloop-agent-pack is a highly structured multi-agent orchestration system for software development, defining six specialized roles (Product, Core Dev, Dev, Test, Marketing, Research) with complex inter-agent communication via `sessions_send`. While the instructions are professional and aligned with the stated purpose, the agents are granted broad, high-risk capabilities including shell access (`Bash`) in `devloop-core-dev.md`, `devloop-dev.md`, and `devloop-test.md`, as well as network access (`WebSearch`, `WebFetch`) in the other agents. The system also utilizes a persistent memory and behavior override mechanism (`SOUL.override.md`, `MEMORY.md`) and references an external `BOOTSTRAP.md` for initialization. Per the analysis criteria, the presence of these risky capabilities, even when plausibly needed for the stated purpose, warrants a suspicious classification.

能力评估

✓ Purpose & Capability

The name/description describe a multi-agent dev workflow and the repository contains 6 agent specs, workflow documentation, and templates. The agent frontmatter lists tools (Read/Write/WebSearch/WebFetch/Bash/etc.) appropriate for the roles (Product, Research use web; Dev/Core/Test require shell/git). No unrelated credentials, binaries, or odd install steps are requested.

ℹ Instruction Scope

Runtime instructions explicitly direct agents to read/write workspace files (SOUL.override.md, SOUL.md, USER.md, memory/*, MEMORY.md, shared/ directories, templates, reports, design/docs). That is consistent with persistence and cross‑agent collaboration, but means agents will load persistent memory and user files (which may contain private context). The SKILL.md also instructs following BOOTSTRAP.md if present (and deleting it afterward) — a customization/initialization mechanism you should inspect.

✓ Install Mechanism

Instruction-only skill with no install spec and no code files to execute on install. No downloads or archive extraction are requested. Low installation risk from external code.

✓ Credentials

The skill declares no required environment variables, credentials, or config paths. Agents may perform network fetches (web_search/web_fetch) but do not request API keys. This aligns with the intended research/marketing/product functions.

⚠ Persistence & Privilege

Agents (Core Dev, Dev, Test) are explicitly allowed to run shell/git operations and Core Dev is permitted to operate on the main branch. Those are high‑privilege capabilities (ability to execute commands and modify source). While coherent for a 'Core Dev' role, they increase risk if you do not trust the skill or its inputs — review and limit these capabilities if you want stricter boundaries.

如何使用

确保已安装 OpenClaw（本地或 Docker 部署）
在对话框中输入安装命令：/install devloop-agent-pack
安装完成后，直接呼叫该 Skill 的名称或使用 /devloop-agent-pack 触发
根据 Skill 的参数说明提供必要输入，即可获得结构化输出

版本历史

v1.0.0

Version 1.0.0 introduces a comprehensive multi-agent workflow for DevLoop: - Adds detailed descriptions of six specialized agents (Product, Core Dev, Dev, Test, Marketing, Research) and their roles in product-driven development. - Documents the complete lifecycle—from product discovery, design, coding, testing, to marketing—supported by stage-by-stage workflows and communication protocols. - Specifies shared file conventions, session startup flows, agent collaboration methods, security guidelines, and memory management standards. - Provides a list of template files for core artifacts (design docs, test specs, reports, etc.) and explains their intended usage. - Includes guidance on customizing agent behavior, troubleshooting collaboration, and referencing in-depth protocol and design background materials.

元数据

Slug devloop-agent-pack

版本 1.0.0

许可证 MIT-0

累计安装 1

当前安装数 1

历史版本数 1

常见问题