Multi-Topic Deep Research Skill

Overview

Intelligent multi-topic deep research tool that automatically analyzes materials and generates systematic research documents. Supports arbitrary material input, launching multiple independent research agents in parallel for deep retrieval, forming a concise research knowledge base.

Core Principles:

• - Only perform information retrieval, summarization, and professional expression transformation
• No adding facts, no fabricating information
• Focus on single problems, concise output, just enough to solve the problem
• Universal design, applicable to legal, business, technical, academic, and other fields

---

Trigger Conditions

Trigger with /multi-search command, or when users request:

• - Deep research on multiple related topics
• Systematic information retrieval
• Multi-perspective analysis integration
• Structured research report generation

---

Input Format

Method 1: File-based

CODEBLOCK0

Method 2: Direct Paste

CODEBLOCK1

Method 3: Specify Topics

CODEBLOCK2

---

Processing Flow

Phase 1: Analysis Preparation

1. 1. Read input materials
2. Extract research topic list

- With clear topics: Use directly - Without clear topics: Auto-extract from materials

1. 3. Topic splitting principles:

- Clear direction: Each topic corresponds to a unique retrieval direction - Avoid overlap: Ensure no duplicate retrieval keywords between topics - Focus on problems: Each topic solves one specific problem

1. 4. Determine project name and output location

Phase 2: Output Directory Detection

Detect project structure by priority:

1. 1. Priority detection: output/ directory -> Use INLINECODE2
2. Secondary detection: Current working directory -> Use INLINECODE3
3. Fallback: User's current directory -> Use INLINECODE4

Create directory: INLINECODE5

Phase 3: Parallel Deep Research

Launch an independent general-purpose research agent for each research topic.

Context Transfer (Main Agent -> Research Agent):

• - Project key information (background, objectives, core problems)
• Complete topic list and retrieval scope for each topic
• Assigned keyword directions (basis for avoiding duplicates)
• Specific requirement background

Deduplication Mechanism:

Each research agent must follow this process before starting retrieval:

1. 1. Pre-retrieval Declaration:

- Declare in current context: "I will search [Keyword A, Keyword B] for researching [Topic Name]" - Wait for main agent confirmation of no duplicates before starting

1. 2. Main Agent Review:

- Check if the agent's declared keywords duplicate assigned directions - If duplicates found, notify the agent to pivot to other directions

1. 3. Dynamic Adjustment:

- If a direction is already covered by other agents, pivot to related but different angles - Record adjusted retrieval directions

Deep Retrieval Requirements:

• - 4-6 rounds of deep retrieval
• Auto-select WebSearch (discovery) or WebFetch (get full content)
• Differentiated keywords, ensuring each agent covers unique angles

Document Generation:

• - Focus on solving a single core problem
• Concise and clear, just enough to solve the problem
• Include key source links
• Directly usable conclusions and recommendations

Phase 4: Integration Output

1. 1. Generate research overview document (000.Research-Overview.md)
2. Integrate core findings from all research agents
3. Create inter-document navigation links
4. Add comprehensive recommendations and immediate action list

---

Output Format

Directory Structure

CODEBLOCK3

Overview Document Format

CODEBLOCK4

Detailed Research Document Format

CODEBLOCK5

---

Link Specifications

Core Principle

All source links must be embedded inline at corresponding positions in the text

CODEBLOCK6

Link Notation Conventions

• - 🔗 -> General web resources
• 📚 -> Academic literature
• 🏛️ -> Institutional websites
• 📄 -> Data sources

---

Document Naming Conventions

Numbering System

• - 00. - Research overview
• INLINECODE8 - Core research
• INLINECODE9 - Important research
• INLINECODE10 - Extended research

Title Guidelines

• - ✅ Use concise titles
• ✅ Avoid special characters
• ✅ Length within 15 words
• ✅ Clearly reflect research subject

---

Quality Standards

Research Agent Quality

• - Focus on single problem: Each research agent solves only one core problem
• Retrieval depth: 4-6 rounds of retrieval (just enough)
• Concise output: Clear and concise, just enough to solve the problem
• Key citations: Cite key sources (just enough)
• Directly usable: Provide directly actionable conclusions and recommendations

Document Quality Standards

• - Clear structure: Chapter titles with clear hierarchy
• Coherent narrative: Paragraph-style narrative, avoid excessive listing
• Accurate links: All links embedded inline at corresponding positions
• Consistent format: Follow unified format specifications
• Strong actionability: Provide specific steps, tools, commands

---

Precautions

Prohibited Actions

• - ❌ Do not create sub-subdirectories (e.g., "reference-materials/")
• ❌ Do not generate separate executive summary files
• ❌ Do not use excessive bullet-point listing format
• ❌ Do not list references separately at document end
• ❌ Do not add redundant progress tracking sections

Recommended Practices

• - ✅ Use narrative paragraph expression
• ✅ Embed links at corresponding text positions
• ✅ Keep research overview concise
• ✅ Provide specific action recommendations
• ✅ Mark clear document numbers

---

Dependencies

This skill relies on Claude Code built-in tools, no additional configuration needed:

• - WebSearch: Search discovery
• WebFetch: Get full content
• Task: Launch independent research agents

---

Changelog

Version	Date	Changes
v1.1.0	2025-03-15	Translated to English
v1.0.0

2025-02-15 | Migrated from Command to Skill, renamed to multi-topic deep research (multi-search) |