Multi Search
Run parallel subagent deep research on distinct angles with keyword dedup and structured output folders for complex topics.
Overview
Multi Search is an agent skill most often used in Idea (also Validate, Build) that coordinates parallel subagents on distinct research angles with keyword dedup and structured report output.
Install
npx skills add https://github.com/cat-xierluo/legal-skills --skill multi-searchWhat is this skill?
- Parallel Subagent architecture with a coordinating main agent
- Dedup gate: subagents declare keywords; coordinator blocks overlapping search directions
- Smart output paths: output/, current directory, or ./research/ fallback
- Differentiated retrieval directions to avoid duplicate sections (v2.3+ principle)
- 12-layer directory structure and unified numbering from prior deep-research lineage
- 12-layer directory structure in unified deep-research lineage
- v1.1.0 adds keyword dedup review before each subagent search
Adoption & trust: 678 installs on skills.sh; 323 GitHub stars; 2/3 security scanners passed (skills.sh audits).
What problem does it solve?
You need deep coverage on a complex topic but one agent keeps re-searching the same angles and duplicating sections.
Who is it for?
Builders running multi-angle desk research (legal, competitive, or technical) who can supervise subagent keyword assignments.
Skip if: Quick single-query lookups, strictly commercial reuse without honoring CC BY-NC-SA 4.0, or teams that cannot run parallel subagents.
When should I use this skill?
You need multi-theme depth research with parallel subagents, declared keywords, and automated output directory detection.
What do I get? / Deliverables
You get a folder of differentiated sub-reports with assigned keywords, reduced overlap, and paths chosen automatically for downstream synthesis.
- Structured research folder with differentiated sub-reports
- Keyword assignment log from coordinator review
Recommended Skills
Journey fit
Spans multiple journey phases - primary shelf plus alternate fits below.
First appears in Idea → research when you need breadth before committing, but the same orchestration supports later validation and doc-heavy build work. Research subphase is the canonical entry for multi-angle retrieval, splitting topics, and assembling markdown research trees.
Where it fits
Split a broad legal or market question into non-overlapping subagent searches before choosing what to build.
Gather multi-source evidence to narrow MVP scope without repeated caselaw or competitor sections.
Produce a layered research directory with unified numbering for product or compliance documentation.
Research differentiated angles for long-form content without duplicating the same citations across drafts.
How it compares
Orchestrated multi-agent research workflow, not a single-shot browser search or static SEO keyword list.
Common Questions / FAQ
Who is multi-search for?
Solo builders and researchers who use agent subagents for serious, multi-angle literature or web research with coordinated outputs.
When should I use multi-search?
In Idea research to map a market or doctrine; in Validate scope to gather evidence; in Build docs when assembling long-form referenced material—whenever breadth needs parallel, non-duplicative searches.
Is multi-search safe to install?
It implies network retrieval and multi-agent execution; review the Security Audits panel on this Prism page and your agent's tool permissions before use.
SKILL.md
READMESKILL.md - Multi Search
# Changelog ## [1.1.0] - 2026-02-15 ### 新增 - **去重检查机制**：每个 Subagent 检索前需声明关键词，主 Agent 审核后确认无重复再开始 - 动态调整：当发现方向重复时，通知 Agent 转向其他角度 - 重命名为通用技能名称 multi-search（多主题深度研究） ### 改进 - 优化上下文传递，增加已分配关键词方向信息 ## [1.0.0] - 2026-02-15 ### 新增 - 从 Command 迁移为 Skill，保留完整功能 - 采用智能输出目录检测机制（优先 output/，其次当前目录，兜底 ./research/） - 保持并行 Subagent 架构，支持完全自动化执行 --- ## 原始版本记录（Command 时期） ### v2.3.0 - 2026-01-05 - **避免内容重复**：添加课题拆分原则，强化差异化检索方向，明确上下文传递机制 ### v2.2.0 - 2026-01-05 - **架构升级**：采用独立Subagent并发架构，聚焦单一问题，精简输出 ### v2.1.0 - 2026-01-05 - 简化统一深度模式，不区分输入来源 ### v2.0.0 - 2026-01-05 - 整合深度研究功能，支持12层目录结构和统一编号 ### v1.2.0 - 2025-11-12 - 增强链接导航和内嵌链接 ### v1.1.0 - 2025-11-12 - 精简版本，保留核心功能 ### v1.0.0 - 2025-11-12 - 初始版本 Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International Copyright (c) 2025 杨卫薪律师（微信ywxlaw） ======================================================================= This work is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License. You are free to: - Share — copy and redistribute the material in any medium or format - Adapt — remix, transform, and build upon the material Under the following terms: - Attribution — You must give appropriate credit, provide a link to the license, and indicate if changes were made. - NonCommercial — You may not use the material for commercial purposes. - ShareAlike — If you remix, transform, or build upon the material, you must distribute your contributions under the same license. No additional restrictions — You may not apply legal terms or technological measures that legally restrict others from doing anything the license permits. To view a copy of this license, visit: http://creativecommons.org/licenses/by-nc-sa/4.0/ ======================================================================= Commercial License For commercial use licenses, please contact: Email: secretxierluo@gmail.com WeChat: ywxlaw (微信) --- name: multi-search homepage: https://github.com/cat-xierluo/legal-skills author: 杨卫薪律师（微信ywxlaw） version: "1.1.0" license: MIT description: 智能多主题深度研究工具，支持任意材料输入，使用独立研究 Agent进行并行深度检索并生成系统化研究文档。本技能应在用户需要深度研究多个相关课题、进行系统性信息检索、整合多角度分析时使用。 --- # 多主题深度研究技能 ## 概述 智能多主题深度研究工具，自动分析材料并生成系统化研究文档。支持任意材料输入，通过**并行启动多个独立研究 Agent**进行深度检索，形成精简的研究知识库。 **核心原则**： - 只做信息检索、归纳与专业表述转化 - 不新增事实，不虚构信息 - 聚焦单一问题，精简输出，能够解决问题即可 - 通用设计，可适用于法律、商业、技术、学术等各领域 --- ## 触发条件 使用 `/multi-search` 命令触发，或当用户请求： - 深度研究多个相关课题 - 进行系统性信息检索 - 整合多角度分析 - 需要生成结构化研究报告 --- ## 输入格式 ### 方式一：基于文件 ``` /multi-search @文档路径.md ``` ### 方式二：直接粘贴 ``` /multi-search [粘贴材料内容] ``` ### 方式三：指定课题 ``` /multi-search 项目：[项目名称] 研究课题： 1. [课题一] 2. [课题二] 3. [课题三] ``` --- ## 处理流程 ### 阶段1：分析准备 1. 读取输入材料 2. 提取研究课题清单 - 有明确课题：直接使用 - 无明确课题：自动从材料中提取 3. 课题拆分原则： - **方向明确**：每个课题对应独特的检索方向 - **避免重叠**：确保课题之间检索关键词不重复 - **聚焦问题**：每个课题解决一个具体问题 4. 确定项目名称和输出位置 ### 阶段2：输出目录智能检测 按优先级检测项目结构： 1. **优先检测**：`output/` 目录 → 使用 `output/[项目名]/` 2. **其次检测**：当前工作目录 → 使用 `./[项目名]/` 3. **兜底方案**：用户当前目录 → 使用 `./research/` 创建目录：`[输出目录]/03 - 🔍 深度研究/` ### 阶段3：并行深度研究 为每个研究课题启动独立的 `general-purpose` 独立研究 Agent。 **上下文传递**（主Agent → 独立研究 Agent）： - 项目关键信息（背景、目标、核心问题） - 完整的课题清单及各课题的检索范围 - 已分配的关键词方向（避免重复的依据） - 具体需求背景 **去重检查机制**： 每个 独立研究 Agent 在开始检索前，必须遵循以下流程： 1. **检索前声明**： - 在当前上下文中声明："我将检索 [关键词A, 关键词B] 用于研究 [课题名称]" - 等待主 Agent 确认无重复后再开始 2. **主 Agent 审核**： - 检查该 Agent 声明的关键词是否与已分配方向重复 - 如发现重复，及时通知该 Agent 转向其他方向 3. **动态调整**： - 如果某方向已被其他 Agent 覆盖，该 Agent 应转向相关但不同的角度 - 记录调整后的检索方向 **深度检索要求**： - 4-6轮深度检索 - 自动选择 WebSearch（搜索发现）或 WebFetch（获取完整内容） - 关键词差异化，确保每个 Agent 覆盖独特角度 **文档生成**： - 聚焦解决单一核心问题 - 简洁明了，能够解决问题即可 - 包含关键来源链接 - 直接可用的结论和建议 ### 阶段4：整合输出 1. 生成研究总览文档（000.研究总览.md） 2. 整合所有 独立研究 Agent 的核心发现 3. 创建文档间导航链接 4. 添加综合建议和立即行动清单 --- ## 输出格式 ### 目录结构 ``` [输出目录]/ └── [项目名]/ └── 03 - 🔍 深度研究/ ├── 000.研究总览.md ├── YYMMDD [研究课题一].md ├── YYMMDD [研究课题二].md └── ... ``` ### 总览文档格式 ```markdown # [项目名称] 深度研究总览 **生成时间**: YYYY-MM-DD **研究方式**: N个独立研究 Agent，各进行4-6轮深度检索 **