solo-swarm
启动 3 个并行研究代理(市场、用户、技术),同时从多个角度研究一个想法。当用户说“群体研究”、“并行研究”、“快速调查”、“3 个代理”、“团队研究”或想要更快地替代 /research 时使用。产生研究.md。请勿用于单独研究(使用/研究)或想法评分(使用/验证)。
安装 / 下载方式
TotalClaw CLI推荐
totalclaw skills install shop:totalclaw:totalclaw~fortunto2-solo-swarmcURL直接下载,无需登录
curl -fsSL https://skills.taituai.com/api/skills/totalclaw%3Atotalclaw~fortunto2-solo-swarm/file -o fortunto2-solo-swarm.md# /swarm
Create an agent team to research "$ARGUMENTS" from multiple perspectives in parallel.
## Team Structure
Spawn 3 teammates, each with a distinct research focus:
### 1. Market Researcher
Focus: competitors, market size, pricing models, business models.
- Search for direct and indirect competitors
- Find market reports with TAM/SAM/SOM figures
- Analyze pricing strategies and monetization
- Identify market gaps and opportunities
- Check Product Hunt, G2, Capterra for existing products
### 2. User Researcher
Focus: pain points, user sentiment, feature requests.
- Search Reddit for user discussions (`site:reddit.com <query>` via WebSearch, or MCP `web_search` if available)
- Search Hacker News for tech community opinions (`site:news.ycombinator.com`)
- If MCP `session_search` available: check if this idea was researched before in past sessions
- Find app reviews and ratings
- Extract direct user quotes about frustrations
- Identify unmet needs and feature requests
### 3. Technical Analyst
Focus: feasibility, tech stack, existing solutions, implementation complexity.
- Search GitHub for open-source alternatives (`site:github.com <query>`)
- Evaluate tech stack options
- If MCP `project_info` available: check existing projects for reusable code
- If MCP `codegraph_explain` available: get architecture overview of similar existing projects
- If MCP `codegraph_query` available: find shared packages across projects
- If MCP `project_code_search` available: search for reusable patterns, services, infrastructure across existing projects
- Assess implementation complexity and timeline
## Search Backends
Teammates should use available search tools:
- **WebSearch** (built-in) — broad discovery, market reports, always available
- **WebFetch** — scrape specific URLs for details, always available
- **MCP `web_search`** (if available) — additional search with engine routing
- **MCP `kb_search`** (if available) — search local knowledge base for related research
**Domain filtering:** use `site:github.com`, `site:reddit.com` etc. for targeted results.
## Coordination
- Each teammate writes findings to a shared task list
- Require plan approval before teammates start deep research
- After all complete, synthesize findings into `research.md`
- Use the research.md format from `/research` skill
## Output
After team completes, the lead should:
1. Synthesize findings from all 3 teammates
2. Write `research.md` to `docs/` in the current project directory
3. Provide GO / NO-GO / PIVOT recommendation
4. Suggest next step: `/validate <idea>`
## Common Issues
### Agent team not available
**Cause:** `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS` env var not set.
**Fix:** Ensure `.claude/settings.json` has `"env": {"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"}`.
### Teammates produce overlapping findings
**Cause:** Research areas not clearly separated.
**Fix:** Each teammate has a distinct focus (market/users/tech). The lead synthesizes and deduplicates findings.
### Web search returns limited results
**Cause:** No additional search backends configured.
**Fix:** Teammates fall back to WebSearch (built-in) which is always available. For richer results with engine routing (Reddit, GitHub, YouTube), set up [SearXNG](https://github.com/fortunto2/searxng-docker-tavily-adapter) (private, self-hosted, free) and configure solograph MCP.
---
## 中文说明
# /swarm
创建一个代理团队,从多个角度并行研究 "$ARGUMENTS"。
## 团队结构
派生 3 名队友,每人有独立的研究侧重点:
### 1. 市场研究员
侧重:竞争对手、市场规模、定价模型、商业模式。
- 搜索直接和间接竞争对手
- 查找带有 TAM/SAM/SOM 数据的市场报告
- 分析定价策略和变现方式
- 识别市场空白与机会
- 在 Product Hunt、G2、Capterra 上查找现有产品
### 2. 用户研究员
侧重:痛点、用户情绪、功能需求。
- 在 Reddit 上搜索用户讨论(通过 WebSearch 使用 `site:reddit.com <query>`,或在可用时使用 MCP `web_search`)
- 在 Hacker News 上搜索技术社区观点(`site:news.ycombinator.com`)
- 如果 MCP `session_search` 可用:检查过去的会话中是否曾研究过这个想法
- 查找应用评论和评分
- 提取用户关于不满的直接引述
- 识别未满足的需求和功能请求
### 3. 技术分析师
侧重:可行性、技术栈、现有解决方案、实现复杂度。
- 在 GitHub 上搜索开源替代方案(`site:github.com <query>`)
- 评估技术栈选项
- 如果 MCP `project_info` 可用:检查现有项目中可复用的代码
- 如果 MCP `codegraph_explain` 可用:获取类似现有项目的架构概览
- 如果 MCP `codegraph_query` 可用:查找跨项目共享的软件包
- 如果 MCP `project_code_search` 可用:在现有项目中搜索可复用的模式、服务和基础设施
- 评估实现复杂度与时间线
## 搜索后端
队友应使用可用的搜索工具:
- **WebSearch**(内置)—— 广泛发现、市场报告,始终可用
- **WebFetch** —— 抓取特定 URL 以获取细节,始终可用
- **MCP `web_search`**(如果可用)—— 带引擎路由的额外搜索
- **MCP `kb_search`**(如果可用)—— 在本地知识库中搜索相关研究
**域名过滤:** 使用 `site:github.com`、`site:reddit.com` 等以获取定向结果。
## 协调
- 每名队友将发现写入共享任务列表
- 在队友开始深入研究之前要求审批计划
- 全部完成后,将发现综合到 `research.md` 中
- 使用 `/research` 技能的 research.md 格式
## 输出
团队完成后,负责人应:
1. 综合所有 3 名队友的发现
2. 将 `research.md` 写入当前项目目录下的 `docs/`
3. 给出 GO / NO-GO / PIVOT 建议
4. 建议下一步:`/validate <idea>`
## 常见问题
### 代理团队不可用
**原因:** 未设置 `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS` 环境变量。
**修复:** 确保 `.claude/settings.json` 中包含 `"env": {"CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"}`。
### 队友产生重叠的发现
**原因:** 研究领域没有清晰区分。
**修复:** 每名队友都有独立的侧重点(市场/用户/技术)。负责人综合并去重各项发现。
### Web 搜索返回的结果有限
**原因:** 未配置额外的搜索后端。
**修复:** 队友回退到始终可用的 WebSearch(内置)。若想获得带引擎路由(Reddit、GitHub、YouTube)的更丰富结果,请搭建 [SearXNG](https://github.com/fortunto2/searxng-docker-tavily-adapter)(私有、自托管、免费)并配置 solograph MCP。