← Back to Skills Marketplace
krislu1221

Claw Roundtable Skill

by Krislu · GitHub ↗ · v2.0.0 · MIT-0
cross-platform ✓ Security Clean
197
Downloads
0
Stars
0
Active Installs
5
Versions
Install in OpenClaw
/install claw-roundtable-skill
Description
多 Agent 深度讨论系统 V2。需求驱动的智能专家匹配系统,模拟真实圆桌会议。 集成 146 个全领域专家(engineering/design/marketing/sales/product 等),支持任何复杂问题的多专家讨论。 核心改进: 1. 需求智能拆解 → 精准匹配专家(从 146 个专家库中选择)...
README (SKILL.md)

RoundTable V2 Skill - 需求驱动的多专家讨论系统

技能说明

RoundTable V2 是一个需求驱动的多专家 Agent 讨论系统。核心理念:

  1. 先拆解需求,再匹配专家 - 不再固定 3-5 个专家
  2. 按议题分治讨论 - 不再固定 5 轮
  3. 排除不相关专家 - 测试专家不参与设计阶段
  4. 动态适配复杂度 - 简单需求快速处理

触发词

  • RoundTable
  • 圆桌会议
  • 圆桌讨论
  • 多 Agent 讨论
  • 多专家讨论
  • 需求分析
  • 方案评审

使用示例

基础用法

RoundTable 讨论一下:智能待办应用的架构设计

指定复杂度

RoundTable 高复杂度:智能待办应用从 0 到 1 完整设计
RoundTable 中复杂度:用户认证模块设计
RoundTable 低复杂度:PR 代码审查

指定专家

RoundTable 指定专家 [engineering, ux_designer]:任务管理界面设计

核心改进(V2 vs V1)

维度 V1(旧版) V2(新版)
需求分析 ❌ 无,直接讨论 ✅ 智能拆解需求
专家匹配 ❌ 固定 3-5 个 ✅ 按需动态选择
讨论流程 ❌ 固定 5 轮 ✅ 按议题分治
专家排除 ❌ 无 ✅ 测试不参与设计
复杂度适配 ❌ 无 ✅ 高/中/低自动适配

需求类型识别

系统自动识别以下需求类型:

类型 关键词 推荐专家
产品定位 产品、功能、用户、需求、定位 产品经理、商业分析师
技术架构 架构、技术栈、后端、前端、数据库 工程专家、架构师
安全合规 安全、认证、授权、加密、隐私 安全工程师、法务
用户体验 体验、界面、交互、设计、UI UX 设计师、UI 设计师
AI/ML AI、智能、算法、模型、推荐 AI 工程师、ML 工程师
性能优化 性能、并发、延迟、优化、缓存 性能工程师、DBA
商业模式 商业、盈利、收入、市场、竞争 商业分析师、营销
数据设计 数据、数据库、表结构、字段 DBA、数据工程师

专家库

技术类

专家 ID 名称 擅长领域
engineering 工程专家 架构、性能、数据
architect 架构师 架构、安全
security_engineer 安全专家 安全、隐私
ai_engineer AI 工程师 AI 功能、模型
ml_engineer ML 工程师 模型训练、优化
data_scientist 数据科学家 数据分析、统计
performance_engineer 性能工程师 性能优化、监控
database_admin 数据库专家 数据库设计、优化
devops DevOps 专家 CI/CD、部署

设计类

专家 ID 名称 擅长领域
ux_designer UX 设计师 用户体验、交互
ui_designer UI 设计师 视觉设计、品牌

产品类

专家 ID 名称 擅长领域
product_manager 产品经理 产品定位、需求
business_analyst 商业分析师 商业模式、市场
marketing 营销专家 增长、品牌
legal 法务专家 合规、法律

测试类(特殊)

专家 ID 名称 擅长领域 排除阶段
qa_engineer 测试专家 性能测试 架构、产品阶段

讨论流程

V2 流程

Step 1: 需求智能拆解
└─ 分析用户输入,识别需求类型

Step 2: 专家精准匹配
└─ 根据需求类型,匹配最相关的专家

Step 3: 用户确认配置
└─ 展示推荐的专家阵容和议题

Step 4: 按议题分治讨论
├─ 议题 1: 技术架构(工程专家主导)
├─ 议题 2: AI 功能(AI 工程师主导)
└─ 议题 3: 用户体验(UX 设计师主导)

Step 5: 整合方案
└─ 将各议题结论整合成完整方案

复杂度适配

复杂度 专家数 议题数 预计耗时 适用场景
2 2 2-5 分钟 简单功能、代码审查
3 3 5-10 分钟 模块设计、功能规划
5 5 15-30 分钟 核心产品、技术选型

实际案例

案例 1:智能待办应用架构设计

输入:RoundTable 讨论一下:智能待办应用的架构设计

Step 1: 需求分析
检测到的需求类型:architecture, ai_ml, ux_design
推荐专家:engineering, ai_engineer, ux_designer
排除专家:qa_engineer(不参与架构阶段)

Step 2: 用户确认
📋 RoundTable V2 配置

讨论主题:智能待办应用的架构设计

推荐专家阵容:
- 工程专家(技术架构)
- AI 工程师(智能功能)
- UX 设计师(用户体验)

关键议题:
- 技术架构 (high)
- AI 功能 (high)
- 用户体验 (medium)

预计耗时:15 分钟
预计 Token:约 40,000

Step 3: 分议题讨论

议题 1: 技术架构(工程专家主导)
→ 结论:React + Node.js + PostgreSQL

议题 2: AI 功能(AI 工程师主导)
→ 结论:本地模型优先 + 云端备份

议题 3: 用户体验(UX 设计师主导)
→ 结论:自然语言输入 + 智能提醒

Step 4: 整合方案
→ 完整的技术架构文档

案例 2:简单功能评审

输入:RoundTable 低复杂度:任务标签功能设计

Step 1: 需求分析
检测到的需求类型:architecture
推荐专家:engineering
复杂度:低 → 最多 2 个专家,2 个议题

Step 2: 快速讨论
议题 1: 数据模型设计
议题 2: API 设计

Step 3: 整合方案
→ 简洁的设计文档

总耗时:3 分钟
Token 消耗:约 10,000

API 参考

快捷函数

# 分析需求
from roundtable_engine_v2 import analyze_requirement

result = analyze_requirement("智能待办应用的架构设计")
print(result)
# {
#     "topic": "智能待办应用的架构设计",
#     "detected_types": ["architecture", "ai_ml"],
#     "recommended_experts": ["engineering", "ai_engineer"],
#     "excluded_experts": ["qa_engineer"],
#     "key_topics": [...]
# }

# 选择专家
from requirement_analyzer import select_experts_for_topic

experts = select_experts_for_topic("智能待办应用的架构设计")
print(experts)  # ["engineering", "ai_engineer", "ux_designer"]

运行 RoundTable

from roundtable_engine_v2 import run_roundtable_v2

# 自动复杂度
await run_roundtable_v2(
    topic="智能待办应用的架构设计",
    mode="pre-ac",
    complexity="auto",  # auto/high/medium/low
    user_channel="user_channel_id"
)

# 指定专家
await run_roundtable_v2(
    topic="任务标签功能设计",
    custom_experts=["engineering", "database_admin"],
    complexity="low"
)

配置选项

复杂度配置

complexity="auto"     # 自动根据需求类型数量判断
complexity="high"     # 高复杂度(5 专家 +5 议题)
complexity="medium"   # 中复杂度(3 专家 +3 议题)
complexity="low"      # 低复杂度(2 专家 +2 议题)

模式配置

mode="pre-ac"   # AC 前讨论(方案设计)
mode="post-ac"  # AC 后审查(代码审查、安全审计)

自定义专家

custom_experts=["engineering", "ux_designer"]  # 指定专家列表

最佳实践

✅ 推荐做法

  1. 明确需求类型 - 在主题中包含关键词(架构/AI/体验等)
  2. 合理选择复杂度 - 简单需求用 low,核心产品用 high
  3. 排除不相关专家 - 设计阶段排除测试专家
  4. 聚焦关键议题 - 不要试图一次讨论所有问题

❌ 避免做法

  1. 过度使用 - 简单问题用 RoundTable(杀鸡用牛刀)
  2. 专家过多 - 超过 5 个专家会导致协调困难
  3. 议题过散 - 一次讨论超过 5 个议题
  4. 强行共识 - 不是所有议题都需要共识

成本对比

场景 V1 成本 V2 成本 节省
高复杂度 100% 100% 0%
中复杂度 100% 50% 50%
低复杂度 100% 25% 75%

整体节省:约 50%(假设高 20% + 中 50% + 低 30%)

作者

虾软 Claw soft

版本历史

  • V2.0.0 (2026-03-21) - 需求驱动的专家匹配系统

    • 新增需求分析器
    • 按议题分治讨论
    • 排除不相关专家
    • 动态适配复杂度
  • V1.0.0 (2026-03-14) - 初始版本

    • 固定 5 轮讨论
    • 固定 3-5 个专家
Usage Guidance
This package appears to implement the roundtable functionality it claims; however: 1) inspect model_selector.py and any code that calls openclaw.tools or other APIs to confirm which endpoints are used and that no secrets are transmitted; 2) verify the bundled expert Markdown files (agency-agents-zh) are present and sane — the loader reads local files and enforces a same-directory check, but supplying a custom AGENCY_AGENTS_PATH could change behavior; 3) review release.sh and any script that writes or executes on your system before running it; 4) prefer explicit configuration (pass a user_models list or keep default local expert pool) rather than enabling automatic environment-driven model selection if you are concerned about accidental network calls; 5) be aware of metadata inconsistencies (version/agent-count) — they suggest sloppy release practices, so audit code paths you care about before use.
Capability Analysis
Type: OpenClaw Skill Name: claw-roundtable-skill Version: 2.0.0 The RoundTable V2 skill bundle is a comprehensive multi-agent discussion framework designed to match user queries with specialized expert personas. The code demonstrates high quality and security awareness, featuring explicit path traversal protections in `agency_agents_loader.py` (restricting file access to the skill directory) and input sanitization in `requirement_analyzer.py` to prevent basic injection attacks. The orchestration logic uses the standard OpenClaw `sessions_spawn` tool for sub-agent management, and no evidence of data exfiltration, unauthorized network activity, or malicious persistence was found across the code or documentation.
Capability Tags
requires-sensitive-credentials
Capability Assessment
Purpose & Capability
The name/description (multi‑agent roundtable with ~146 experts) aligns with the files included: requirement_analyzer, roundtable_engine_v2, agency_agents_loader, etc. Some metadata/content inconsistencies exist (manifest/version strings and agent-count comments differ in places — e.g., 146 vs 604, __version__ 0.9.1 vs registry 2.0.0). These are sloppy but not in themselves malicious; they reduce confidence in release hygiene.
Instruction Scope
SKILL.md and API examples restrict operations to analyzing a topic, selecting experts, and running the roundtable engine. The runtime files focus on reading bundled agent Markdown files and choosing models. I did not find instructions that ask the agent to read unrelated system credentials, remote arbitrary endpoints, or exfiltrate data. Some code paths reference environment variables (AGENCY_AGENTS_PATH, ROUNDTable_MODELS) for configuration — those are reasonable for this use case but worth reviewing.
Install Mechanism
There is no install spec (instruction-only at packaging level) and the repository files are provided. No network downloads or external installers are defined in the package. A release.sh script exists (not shown in full) — it is not automatically run by an install spec, but you should inspect it before running manually.
Credentials
The skill declares no required env vars or credentials. The code optionally reads AGENCY_AGENTS_PATH and ROUNDTable_MODELS for configuration; these are non‑sensitive config-ish variables and appropriate for the purpose. MODEL_CONFIG claims it will not read apiKey/baseUrl/etc. — if you plan to run this, verify model_selector.py to ensure it does not attempt to consume secrets or external API keys. Also note AgencyAgentsLoader will accept a base path from env but validates it must be inside the skill directory (mitigating arbitrary-file access).
Persistence & Privilege
always:false (normal) and disable-model-invocation:false (normal). The skill does not request system‑wide persistent privileges. It does include functions to export/import model configuration (writing to user paths) — a normal feature but one you should review if you want to audit what gets written.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install claw-roundtable-skill
  3. After installation, invoke the skill by name or use /claw-roundtable-skill
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v2.0.0
RoundTable Skill V2.0.0 — Major platform overhaul introducing demand-driven expert matching and flexible discussion flow. - Added requirement analyzer and new roundtable V2 engine for intelligent demand parsing and expert selection. - Discussions now organized by dynamically identified topics, not fixed rounds; experts assigned per-issue. - Excludes irrelevant experts and adapts flow to task complexity (low/medium/high). - Expanded API and documentation, significant refactor with 24 new and 6 updated files. - Dramatically reduces resource usage for simple/moderate requests; improves relevance and efficiency for all cases.
v0.9.3
Version 0.9.3 Changelog - Expanded skill description to clarify automatic sub-agent creation based on project needs and included detailed behavior about multi-role, multi-model support. - Added explanation of the benefits for users with multiple models and noted increased token and time consumption. - Noted the inclusion of 146 expert prompts for specialized sub-agent thinking. - Clarified that users can contribute opinions during discussions for enhanced results. - Described suitability for brainstorming and project compliance reviews.
v0.9.2
**Major cleanup and simplification in v0.9.2:** - Removed documentation, prompt, requirements, and template files to streamline the codebase. - Added `agent_selector.py` to support agent selection functionality. - The skill’s documentation (`SKILL.md`) now only includes basic descriptions, trigger words, authorship, and run-time behaviors. - Confirmation is now automatic; manual confirmation is no longer required before starting a discussion.
v0.9.1
## v0.9.1 - Added LICENSE file to the project.
v0.9.0
RoundTable Skill 0.9.0 – Enhanced via Real-World Testing - 5-stage discussion process (R1-R5) now fully implemented - Enforced deep critical review (min. 5 risks, 3 defects analysis) - Real-time plan evolution with clear change tracking in R4 - Mandatory resolution of disagreements in R5 - Output includes actionable weekly plans and risk contingency - Improved trigger word recognition - Real sub-agent invocation and per-round context injection for richer discussions
Metadata
Slug claw-roundtable-skill
Version 2.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 5
Frequently Asked Questions

What is Claw Roundtable Skill?

多 Agent 深度讨论系统 V2。需求驱动的智能专家匹配系统,模拟真实圆桌会议。 集成 146 个全领域专家(engineering/design/marketing/sales/product 等),支持任何复杂问题的多专家讨论。 核心改进: 1. 需求智能拆解 → 精准匹配专家(从 146 个专家库中选择)... It is an AI Agent Skill for Claude Code / OpenClaw, with 197 downloads so far.

How do I install Claw Roundtable Skill?

Run "/install claw-roundtable-skill" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Claw Roundtable Skill free?

Yes, Claw Roundtable Skill is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Claw Roundtable Skill support?

Claw Roundtable Skill is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Claw Roundtable Skill?

It is built and maintained by Krislu (@krislu1221); the current version is v2.0.0.

💬 Comments