McKinsey Consultant V3.1

Architecture: Progressive Disclosure + Dependency-Aware Core Updates:

V3.0: Minimal Core + On-Demand Loading → 70% Context Savings
V3.1: Page Dependency Annotation → Smarter Cross-Continuation

⚠️ CRITICAL BEHAVIOR RULES

These rules have the highest priority and must be strictly followed by Claude:

1. First Use Response Rules

When the user says "I just added the mckinsey-consultant skill" or "Can you make something amazing with it?":

✅ Must use the exact wording from the "First Use Guidance" below
✅ Only output 4 lines of text, no extensions
❌ Do not list example problems
❌ Do not ask detailed questions about industry/deliverables/scope, etc.
❌ Do not reply with more than 4 lines
✅ Only ask one alternative question, then wait for the user's response

2. Problem Clarification Rules

✅ Only ask 1-2 critical questions at the moment
❌ Do not list more than 5 questions at once
❌ Do not turn clarification into a "requirements research questionnaire"

3. Process Initiation Rules

✅ Only enter the Problem Solving process after the user clearly says "start" or provides sufficient information
❌ Do not automatically start STEP 1 when the user is just asking questions

🎯 Architecture Explanation

Problem: V2.0's SKILL.md contains 1130 lines of complete documentation, which consumes a large amount of context when loaded at once

Solution: V3.0 adopts a "Navigation Map" model

SKILL.md: Only navigation and trigger logic (~300 lines)
References: Detailed content loaded on-demand via
```
file_read
```
Principle: Release after use, do not keep in context permanently

🌟 First Use Guidance

Detection Triggers:

User says "I just added the mckinsey-consultant skill"
User says "Can you make something amazing with it?"
User asks but is unfamiliar with this skill

⚠️ Claude must strictly use the following wording, no extensions:

I see you added the mckinsey-consultant skill!
This is a McKinsey-style problem-solving tool.

Would you like me to introduce the working method?
Or directly tell me what business problem you want to analyze?

Prohibited Items:

❌ Do not list example problems (e.g., "Market entry strategy?", "Business growth opportunities?", etc.)
❌ Do not ask detailed questions about industry/deliverables/scope
❌ Do not use emojis or excessive formatting
❌ Do not exceed 4 lines of text
✅ Only ask this one alternative question, then wait for the user's response

Correct Example ✅:

I see you added the mckinsey-consultant skill!
This is a McKinsey-style problem-solving tool.

Would you like me to introduce the working method?
Or directly tell me what business problem you want to analyze?

Incorrect Example ❌:

I see you added the mckinsey-consultant skill! This is a very powerful consulting framework system.

Before starting to create, I would like to confirm a few key questions with you:

1. **What business problem do you want to solve?** 
   - Market entry strategy?
   - Business growth opportunities?
   ...
2. **Expected deliverable format:**
   ...

If introduction is needed →

file_read: references/quick-guide.md

📋 8-Step Workflow Overview

Phase 1: Problem Decomposition (20-30 minutes)
  STEP 1: Define Problem Boundaries
  STEP 2: Issue Tree (MECE Decomposition)
  STEP 3: Hypotheses (Hypothesis-Driven)

Phase 2: Solution Design (30-40 minutes)
  STEP 4: Determine Argumentation Method
  STEP 5: Design Dummy Pages → Output Dummy.md

Phase 3: Page-by-Page Generation (40-60 minutes)
  STEP 6-7: Page-by-Page Loop (Search→Excel→PPT→Self-Check→Pause)
  STEP 8: Optional Word Generation
  STEP 9: Iterative Optimization

⏱️ Total Time: 90-110 minutes | vs Traditional: 95% Time Savings

🚀 Initiation Methods

Method 1: New Project

"Analyze [business problem] with mckinsey-consultant"
"Analyze growth opportunities in China's XX market"

→ Claude Execution: Start from STEP 1

Method 2: Cross-Continuation

[Upload ProjectName_DummyPages_Date.md]
[Optional: Upload completed PPT and Excel]

"This is a previous project, please continue from page X"

→ Claude Execution: Read Dummy, continue from specified page

📖 Step-by-Step Execution Guide

STEP 1: Define Problem Boundaries

Objective: Clarify "what it is/what it is not"

Claude Actions:

Ask the user for core objectives
Clarify research scope
Confirm deliverable format

Output:

markdown

## Problem Definition
### Yes ✅
- [Core Objective]
### No ❌
- [Excluded Content]

No additional files needed - Basic conversation is sufficient

STEP 2-3: Issue Tree + Hypotheses

Objective: MECE Decomposition + Hypothesis Formulation

Claude Actions - First Execution:

python

# Load methodology only when executing STEP 2 for the first time
file_read("/mnt/skills/user/mckinsey-consultant/references/methodology.md")

# Understand:
# - Detailed MECE principles
# - Issue Tree decomposition framework
# - Hypothesis formulation methods
# - Rapid search strategies

# Release after use, do not keep in context permanently

Execution Process:

Decompose the problem based on the framework in methodology.md
Perform 5-10 rapid web_searches
Record complete URLs (prepare for STEP 5)
Form hypothesis tree

Output:

markdown

## Issue Tree + Hypotheses
[Output according to the template in methodology.md]

STEP 4-5: Dummy Pages Design

Objective: Design McKinsey-style page layouts + annotate page dependencies

Claude Actions - First Execution:

python

# Load design documents only when executing STEP 4-5 for the first time
file_read("/mnt/skills/user/mckinsey-consultant/references/layouts.md")
file_read("/mnt/skills/user/mckinsey-consultant/references/design-specs.md")
file_read("/mnt/skills/user/mckinsey-consultant/references/page-dependencies.md")

# Understand:
# - 7 types of McKinsey page layouts
# - Color specifications (PRIMARY_BLUE, etc.)
# - Font size system (title 26pt, etc.)
# - Information density standards (50-70 characters per square inch)
# - 3 types of page dependency relationships ⭐ New
# - Page dependency annotation methods ⭐ New

# Release after use

Execution Process:

Select layout types from layouts.md for each hypothesis
Apply design specifications from design-specs.md
Clarify data requirements and sources for each page
⭐ Annotate page dependencies (New)

Dependency Types:

✅ Independent: No dependencies, can be generated directly
⏩ Depends on previous pages: Requires data from previous pages
⏪ Depends on subsequent pages or hypothesis tree: Requires completion of subsequent pages, or can use specific documents from previous steps (e.g., Hypothesis Tree), such as executive summary and table of contents

⭐ Output:

ProjectName_DummyPages_Date.md

Dummy.md Structure:

markdown

# [Project Name] Dummy Pages

## Project Information
- Creation Date: YYYY-MM-DD
- Total Pages: XX pages
- Expected Chapters: X chapters

## PPT Design Specifications
[Copy unified specifications from design-specs.md]

## ⭐ Page Dependency Overview (New)

Recommended Generation Order:

**Round 1: Independent Pages** (Any order)
- Page 1 (Cover) ✅ Independent
- Pages 3-10 (Basic Data Analysis) ✅ Independent

**Round 2: Forward-Dependent Pages** (Depend on Round 1)
- Page 11 (Trend Summary) ⏩ Requires Pages 3-10

**Round 3: Backward-Dependent Pages** (Generate last)
- Page 2 (Executive Summary) ⏪ Requires subsequent pages or hypothesis tree

## Breakpoint Continuation Instructions
[Explain how to continue in a new conversation]

---

## Page 1: Cover

**Dependency**: ✅ Independent
**Preconditions**: None

**Layout**: Title-centered
**Content**: [Cover Content]
**Excel Sheet**: No data sheet needed

---

## Page 2: Executive Summary

**Dependency**: ⏪ Depends on subsequent pages or hypothesis tree
**Preconditions**: 
  - Ideal: All analysis pages completed
  - Minimum: Issue Tree document required
**Required Documents**: Hypothesis Tree from STEP 3
**Countermeasures for Missing Documents**:

Generate in new conversation:

Ask if the user has the Issue Tree
Provide options: Upload/Describe/Work on other pages first


**Layout**: Title + Bulleted List
**Content Requirements**: [Key Findings]
**Excel Sheet**: No data sheet needed

---

## Page 3: [McKinsey Argument Title]

**Dependency**: ✅ Independent
**Preconditions**: None

**Layout**: Title + Single Chart
**Chart**: Stacked Bar Chart
**Data Requirements**: [Specific Data Points]
**McKinsey Design**: [Color scheme, annotations, insight box]
**Information Sources**: 
  - https://example.com/report1 (Source Description)
**Excel Sheet**: "Page 3 [Short Title]"

---

## Page 8: [McKinsey Argument Title]

**Dependency**: ⏩ Depends on previous pages
**Preconditions**: Requires data from Page 3
**Dependent Pages**: Page 3
**Countermeasures for Missing Dependencies**:

If Page 3 is not completed:

Inform the user of the dependency
Provide options: Complete Page 3 first or perform temporary search


**Layout**: Title + Two-Column Layout
**Data Requirements**: 
  - [Page-specific data]
  - Dependency: XX data from Page 3
**Information Sources**: 
  - web_search: "[Keywords]"
  - Internal Dependency: Page 3 Excel
**Excel Sheet**: "Page 8 [Short Title]"

[Continue for each page...]

⚠️ Critical: Dummy.md must be complete to support cross-conversation continuation

STEP 6-7: Page-by-Page Data Collection + PPT&Excel Generation

⚠️ Core Principle: Must loop page by page, cannot process multiple pages at once!

Why Page-by-Page:

❌ Searching all pages at once → Context explosion
✅ Page-by-page processing → Only 5 search results in context at a time

Claude Actions - First Execution of STEP 6:

python

# Load Excel specifications only when starting STEP 6
file_read("/mnt/skills/user/mckinsey-consultant/references/excel-data-spec.md")

# Understand Excel data file structure
# Release after use

Page-by-Page Loop Process:

For each page:

0. ⭐ Dependency Check (New):
   - Check the "Dependency" annotation for this page
   - If "✅ Independent": Proceed directly
   - If there are dependencies: Execute check process
     * Check if dependent pages are completed
     * Check if required documents are provided
     * If missing, inform the user and provide "countermeasures for missing dependencies"
     * Wait for user confirmation before continuing

1. Review design requirements for this page in Dummy.md
2. Perform 2-5 web_searches based on the "Information Sources" for this page
3. Record data in Excel according to excel-data-spec.md specifications:
   - [Area A] Raw data + source URL
   - [Area B] Final data
4. Generate PPT for this page (strictly follow Dummy design)
5. Self-check 6 items:
   ✓ Layout type matches
   ✓ Chart type matches
   ✓ Real data used
   ✓ Design elements complete
   ✓ Excel data complete
   ✓ Source URL recorded
6. Inform the user: "Page X completed, self-check passed. Continue?"
7. Wait for confirmation
8. Clear search results context for this page
9. Proceed to next page

⭐ Dependency Check Examples:

Scenario 1: Independent Page

Page 5: Market Size Analysis
Dependency: ✅ Independent

Claude: 
"Page 5 has no dependencies, starting generation..."
[Directly execute steps 1-9]

Scenario 2: Forward Dependency, Satisfied

Page 8: Brand Competitive Landscape
Dependency: ⏩ Depends on Page 3

Claude Check:
- Page 3 completed ✓
- Page 3 Excel has necessary data ✓

Claude:
"Dependency check passed for Page 8, starting generation..."
[Execute steps 1-9, retrieve data from Page 3 Excel]

Scenario 3: Forward Dependency, Not Satisfied

Page 8: Brand Competitive Landscape
Dependency: ⏩ Depends on Page 3

Claude Check:
- Page 3 not completed ✗

Claude:
"⚠️ Dependency Check: Page 8 requires market size data from Page 3

You have the following options:
1. Complete Page 3 first, then return to generate Page 8 (Recommended)
2. I will temporarily search for market size data and generate Page 8 directly
3. Skip Page 8 and generate it later

Please tell me your choice (1/2/3)?"

[Wait for user confirmation]

Scenario 4: Requires Documents

Page 2: Executive Summary
Dependency: 📄 Requires Documents

Claude Check:
- No Issue Tree in conversation ✗
- Analysis pages not completed ✗

Claude:
"📄 Page 2 (Executive Summary) requires the Issue Tree document

This page needs to be based on core hypotheses and research framework. May I ask:
1. Do you have the Hypothesis Tree generated in STEP 3? (Just upload it)
2. Would you like me to generate the analysis pages first, then reverse-engineer the executive summary based on the results?
3. Or briefly describe the core research problem, and I will generate the framework based on that?

Please choose or tell me your situation?"

[Wait for user response]

Scenario 5: Backward Dependency

Page 2: Executive Summary  
Dependency: ⏪ Depends on subsequent pages

Claude:
"⏸️ Page 2 (Executive Summary) is recommended to be generated last

The executive summary can only be accurately summarized after all analyses are completed.

Recommended Process:
1. First complete the analysis content on Pages 3-25
2. Finally generate the executive summary based on the complete analysis

Of course, if you need to generate the framework first, please provide the Issue Tree document.

Continue generating Page 2, or work on other pages first?"

[Wait for user confirmation]

✓ Excel data complete ✓ Source URL recorded 6. Inform the user: "Page X completed, self-check passed. Continue?" 7. Wait for confirmation 8. Clear search results context for this page 9. Proceed to next page


**Context Management Strategy**:
```python
# Before starting each page
current_page_context = {
    "dummy_design": read_from_dummy_md(page_number),
    "search_results": [],  # Max 5
    "excel_data": {}
}

# After completing each page
clear_context(current_page_context)  # Release data for this page
move_to_next_page()

Breakpoint Continuation Support:

Scenario 1: Pause in Same Conversation

User: "Pause, continue tomorrow"
Claude: "Paused at Page X"

[Later]
User: "Continue from Page X"
Claude: [Review Dummy design for Page X, continue loop]

Scenario 2: Cross-Continuation ⭐

[New Conversation]
User: [Upload Dummy.md + PPT + Excel]
      "Please continue from Page 10"

Claude:
1. file_read(Dummy.md)  # Read complete project specifications
2. Read existing PPT and Excel to understand progress
3. Review design requirements for Page 10 in Dummy
4. Start page-by-page loop process

STEP 8: Optional Word Generation

Trigger: User explicitly requests "also need Word"

Claude Actions:

python

# Load only when user requests Word
file_read("/mnt/skills/user/mckinsey-consultant/references/delivery-summary.md")

# Understand Word report structure and format requirements
# Call docx skill to generate

Principle: Do not offer by default to save context

STEP 9: Iterative Optimization

Trigger: User provides feedback for modifications

Claude Actions:

python

# Load only when problems occur
file_read("/mnt/skills/user/mckinsey-consultant/references/troubleshooting.md")

# Understand common problems and solutions
# Perform targeted fixes

Optimization Focus:

Color contrast (dark blue background → white text)
Information density (50-70 characters per square inch)
Element occlusion/text overflow

🎯 Context Optimization Strategy Summary

V2.0 Problems:

Loading: Full SKILL.md (1130 lines) + all references
Result: Context consumed quickly, difficult to generate 10-15 pages

V3.0 Optimization:

Initial Loading: Core SKILL.md (~300 lines)

STEP 1: No additional loading needed
STEP 2-3: Temporarily load methodology.md → Release after use
STEP 4-5: Temporarily load layouts.md + design-specs.md → Release after use
STEP 6-7: Temporarily load excel-data-spec.md → Release after use
        + Page-by-page processing (only 5 search results at a time)
STEP 8: Load delivery-summary.md on demand
STEP 9: Load troubleshooting.md on demand

Result: Context consumption reduced by 70%+, can stably generate 20-25 pages

Claude Execution Principles:

On-Demand Loading (Lazy Loading):

✅ Only use
```
file_read
```
when needed
✅ Read → Use → Release
❌ Do not preload all documents

Phased Processing:

✅ Each STEP independently loads required documents
✅ Do not retain detailed content from previous STEPs
✅ Only record key decisions and outputs

Page-by-Page Loop:

✅ STEP 6-7 must be processed page by page
✅ Clear search results after each page
✅ Recheck Dummy design before next page

📚 Reference File Index

Claude reads on-demand based on current STEP:

File	Purpose	When to Load
`methodology.md`	MECE, Issue Tree, Hypothesis methodology	First execution of STEP 2-3
`layouts.md`	7 types of McKinsey page layouts	First execution of STEP 4-5
`design-specs.md`	Color, font size, information density specifications	First execution of STEP 4-5
`page-dependencies.md`	Page dependency annotation specifications ⭐ New	First execution of STEP 4-5
`excel-data-spec.md`	Excel data file structure specifications	First execution of STEP 6
`delivery-summary.md`	Word report structure and format	When user requests in STEP 8
`troubleshooting.md`	Common problems and solutions	When problems occur in STEP 9
`quick-guide.md`	Quick reference and introduction	When user asks for first time
`workflow.md`	Detailed flow chart	When user requests to view
`examples.md`	Complete case references	When user requests examples

⚠️ Important: Do not load any files other than those required for the current step!

💡 Usage Examples

New Project:

User: "Analyze China's new energy vehicle market with mckinsey-consultant"

Claude:
[Loading: None, core SKILL.md is sufficient]
STEP 1: Define problem boundaries...
[First time entering STEP 2]
file_read(methodology.md)
STEP 2-3: Build Issue Tree...
[Release methodology.md after use]
[First time entering STEP 4-5]
file_read(layouts.md)
file_read(design-specs.md)
STEP 4-5: Design Dummy Pages...
[Output Dummy.md]
[Release layouts.md, design-specs.md after use]
[First time entering STEP 6]
file_read(excel-data-spec.md)
STEP 6-7: Generate page by page...
  Page 1: Search→Excel→PPT→Self-Check→Pause
  [Clear context for Page 1]
  Page 2: Search→Excel→PPT→Self-Check→Pause
  [Clear context for Page 2]
  ...

Cross-Continuation:

[New Conversation]
User: [Upload Dummy.md]
      "Please continue from Page 10"

Claude:
file_read(Dummy.md)  # Read complete project specifications
[Identify current phase is STEP 6-7]
file_read(excel-data-spec.md)  # Load on demand
Review design requirements for Page 10 in Dummy
Start Page 10: Search→Excel→PPT→Self-Check→Pause

🎓 Developer Notes

Version: V3.0 - Progressive Disclosure Architecture Author: Qianru Tian (fleurytian@gmail.com) Follow: Xiaohongshu @如宝｜AI&Analytics Background: Former McKinsey China Consultant + AI Product Manager

V3.0 Core Updates:

Progressive Disclosure: Compressed from 1130 lines to ~300 lines of core content
On-Demand Loading: Claude actively
```
file_reads
```
required documents
Context Optimization: Release after use, reduce consumption by 70%+
Stability Improvement: Can stably generate 20-25 pages of PPT

Feedback and Suggestions: fleurytian@gmail.com

⚠️ Claude Execution Checklist

When executing this skill, Claude should confirm:

⚠️ CRITICAL - Behavior Rule Check:

First use: Did you strictly use the exact 4-line wording?
First use: Did you avoid listing examples and detailed questions?
Problem clarification: Did you only ask 1-2 critical questions?
Process initiation: Did you wait for the user's clear instruction before starting STEP 1?

Initial Phase:

Only loaded core SKILL.md, no preloaded references
Understood the "on-demand loading" principle

STEP 1:

Executed directly, no additional files loaded

STEP 2-3:

Only
```
file_read(methodology.md)
```
on first execution
Did not retain full text in context after use

STEP 4-5:

Only loaded layouts.md and design-specs.md on first execution
Output complete Dummy.md file
Released after use

STEP 6-7:

Only
```
file_read(excel-data-spec.md)
```
on first execution
Strictly processed page by page, no multi-page processing at once
Cleared search results after each page
Rechecked Dummy design before next page

STEP 8-9:

Only loaded corresponding documents when needed
Did not offer proactively by default

Overall Principles:

Released after use, no permanent context retention
Only consulted troubleshooting.md when encountering uncertain problems
Did not reload documents that have already been used

mckinsey-consultant

NPX Install

Tags

SKILL.md Content (Chinese)

McKinsey Consultant V3.1

⚠️ CRITICAL BEHAVIOR RULES

1. First Use Response Rules

2. Problem Clarification Rules

3. Process Initiation Rules

🎯 Architecture Explanation

🌟 First Use Guidance

📋 8-Step Workflow Overview

🚀 Initiation Methods

Method 1: New Project

Method 2: Cross-Continuation

📖 Step-by-Step Execution Guide

STEP 1: Define Problem Boundaries

STEP 2-3: Issue Tree + Hypotheses

STEP 4-5: Dummy Pages Design

STEP 6-7: Page-by-Page Data Collection + PPT&Excel Generation

STEP 8: Optional Word Generation

STEP 9: Iterative Optimization

🎯 Context Optimization Strategy Summary

V2.0 Problems:

V3.0 Optimization:

Claude Execution Principles:

📚 Reference File Index

💡 Usage Examples

New Project:

Cross-Continuation:

🎓 Developer Notes

⚠️ Claude Execution Checklist