Topics

#ai-tools #anthropic #bcg #business-frameworks #claude #claude-code #claude-skills #consulting #frameworks #issue-trees #llm #mckinsey #mece #pareto #prompt-engineering #prompts #strategy

Language breakdown

Python 100.0%

▸ Files click to expand

README

Strategy Consultant — 5 Consulting Frameworks

Strategy Consultant — whiteboard-style consulting frameworks for any LLM

Repo: Strategy-Consultant-5-Consulting-Frameworks

Turn any LLM into a Tier-1 strategy consultant who breaks down complex business problems on a whiteboard — MECE, issue trees, hypothesis tests, the Pareto principle, and the "So What?" test, in sequence, every time.

The idea: most LLMs answer business questions in flowing prose that sounds smart but doesn't actually structure your thinking. This repo gives Claude (or any LLM) a strict visual contract: every business problem produces exactly five sections — a MECE breakdown, an ASCII issue tree, a hypothesis-test table, a Pareto blockquote, and a Process/Result/Insight closer. Whiteboard-style, every time. The frameworks aren't new — they're standard McKinsey/BCG/Bain practice — but baking them into a deterministic skill format means you stop getting fluffy answers and start getting consultant-grade ones.

The repo ships several flavors. A sequential master that works in any LLM (Claude.ai, ChatGPT, Gemini, anywhere you can paste a system prompt). A sub-agent master for Claude Code or Cowork environments that support parallel agent delegation. A 7-framework variant that adds 5 Whys and Pyramid Principle on top of the canonical five for high-stakes work. And three industry packs (SaaS, retail, healthcare) with industry-aware vocabulary and root-cause priors. See the Master Variants table below for which to pick.

What you give it, what it gives back

You give it: "Our SaaS churn jumped 4 points last quarter and nobody knows why."

It gives you back, in one response:

MECE breakdown — every plausible churn driver, organized so nothing overlaps and nothing is missed
Issue Tree — an ASCII tree decomposing churn into testable root causes
Hypothesis Test — a table comparing what you'd expect to see if your hypothesis is true vs. what the data actually shows (with at least one control row designed to invalidate the hypothesis)
Pareto Focus — the 20% of drivers responsible for 80% of the churn, called out in a blockquote, with the deprioritized 80% explicitly named
So What? — Process / Result / Insight, with an action specific enough to assign to a person with a deadline

See it in action. Browse examples/showcase/ for 10 real outputs across industries — every one produced by an agent reading a SKILL.md and applying it to a fresh business problem. Or read the canonical end-to-end walk-through: examples/example-analysis.md.

Quick start

Claude Code / Cowork on macOS or Linux

git clone https://github.com/ConrayGambit/Strategy-Consultant-5-Consulting-Frameworks.git
cd Strategy-Consultant-5-Consulting-Frameworks
mkdir -p ~/.claude/skills
cp -r skills/* ~/.claude/skills/

Restart Claude Code. Try /strategy-consultant and paste in a business problem.

Claude Code / Cowork on Windows (PowerShell)

git clone https://github.com/ConrayGambit/Strategy-Consultant-5-Consulting-Frameworks.git
cd Strategy-Consultant-5-Consulting-Frameworks
$skillsPath = "$env:USERPROFILE\.claude\skills"
if (-not (Test-Path $skillsPath)) { New-Item -ItemType Directory -Path $skillsPath -Force | Out-Null }
Copy-Item -Path .\skills\* -Destination $skillsPath -Recurse -Force

Anywhere else (Claude.ai, ChatGPT, Gemini, local models)

Open prompts/strategy-consultant.md, copy the whole file, paste into your LLM as a system prompt or first message, then ask your business question.

For Custom GPTs and Claude Projects, paste it into the "Instructions" field.

Full per-platform install (including Cowork on Windows, ChatGPT Custom GPT setup, etc.) is in INSTALLATION.md.

After installing, verify everything works using TESTING.md — it includes a canonical test prompt and pass criteria for each install path.

Repository layout

Strategy-Consultant-5-Consulting-Frameworks/
├── skills/                                # Anthropic Skills format (SKILL.md + frontmatter)
│   ├── strategy-consultant/               # Master — 5 frameworks, sequential
│   ├── strategy-consultant-agents/        # Master — 5 frameworks, parallel sub-agents
│   ├── strategy-consultant-7/             # Master — 7 frameworks (adds 5 Whys + Pyramid Principle)
│   ├── mece-framework/                    # Sub-skill: MECE
│   ├── issue-trees/                       # Sub-skill: Issue Trees
│   ├── hypothesis-driven/                 # Sub-skill: Hypothesis-Driven
│   ├── pareto-principle/                  # Sub-skill: Pareto / 80-20
│   ├── so-what-test/                      # Sub-skill: So What?
│   ├── five-whys/                         # Sub-skill: 5 Whys
│   └── pyramid-principle/                 # Sub-skill: Pyramid Principle
├── industry-packs/                        # Industry-tailored master variants
│   ├── saas/SKILL.md                      # sc-saas
│   ├── retail/SKILL.md                    # sc-retail
│   └── healthcare/SKILL.md                # sc-healthcare
├── prompts/                               # LLM-agnostic plain prompts (one per skill)
├── examples/
│   ├── example-analysis.md                # Featured walk-through
│   ├── full-runs/                         # 5 industries: retail, nonprofit, manufacturing, government, logistics
│   ├── single-frameworks/                 # 5 standalone-skill demos in different industries
│   └── showcase/                          # 10 real outputs from end-to-end skill testing
├── evals/                                 # Automated structural scorer + manual rubric
│   ├── score.py                           # Python scorer (no dependencies)
│   ├── RUBRIC.md                          # Manual content-quality rubric
│   └── fixtures/                          # Known-bad samples for regression testing
├── INSTALLATION.md                        # Per-platform install guide
├── TESTING.md                             # Canonical test prompt + per-platform pass criteria
├── CONTRIBUTING.md                        # How to add frameworks, examples, industry packs
├── GITHUB-SETUP.md                        # Notes for forkers
├── LICENSE                                # MIT
└── README.md

How it works

The repo only really has three things that matter:

skills/<skill-name>/SKILL.md — Anthropic Skills format with YAML frontmatter (name, description) and the actual instructions in markdown. Claude Code and Cowork load these automatically.
prompts/<skill-name>.md — A condensed, paste-ready version of each skill: it keeps the full output contract (exact section headers, formats, defaults, the control-row rule) but drops the frontmatter, worked examples, and meta sections. For pasting into ChatGPT, Claude.ai, Gemini, custom GPTs, etc.
examples/ — Real input/output pairs you can read to calibrate what good output looks like.

The three master skills (strategy-consultant, strategy-consultant-agents, and strategy-consultant-7) are self-contained — they walk through all five (or all seven) frameworks in one response. The seven sub-skills (mece-framework, issue-trees, hypothesis-driven, pareto-principle, so-what-test, five-whys, pyramid-principle) are alternatives for when you only want one framework. The industry packs (sc-saas, sc-retail, sc-healthcare) are alternative masters with industry-aware MECE defaults and vocabulary. Each also ships a paste-ready prompt in prompts/.

Master variants

The repo ships several master skills. Pick the one that fits your need:

| Variant | Frameworks | When to use | |---|---|---| | strategy-consultant (sequential) | 5 — MECE, Issue Tree, Hypothesis, Pareto, So What? | Default. Works in any LLM. One pass, single response. | | strategy-consultant-agents (sub-agents) | 5 — same | Claude Code / Cowork only. Delegates each framework to a parallel sub-agent. Best for high-stakes complex problems. | | strategy-consultant-7 | 7 — adds 5 Whys + Pyramid Principle | High-stakes work that needs both deeper root-cause drilling AND executive-ready communication structure. | | sc-saas (details) | 5 — same | SaaS-specific problems. Industry-aware MECE defaults (NRR, churn, pipeline), native vocabulary, common root-cause priors. | | sc-retail (details) | 5 — same | Retail / hospitality / multi-unit problems. Comp store, foot traffic, basket, mix. | | sc-healthcare (details) | 5 — same | Hospital ops / clinical operations. Readmissions, ALOS, throughput, access. | | sc-fintech (details) | 5 — same | Fintech: payments / lending / fraud / regulatory. Approval/decline, chargeback, delinquency, KYC. | | sc-manufacturing (details) | 5 — same | Manufacturing / industrial ops. OEE, defect rate, yield, changeover, throughput. | | sc-nonprofit (details) | 5 — same | Nonprofit: fundraising, donor retention, gift mix, program effectiveness. | | sc-government (details) | 5 — same | Public sector / civic ops. Backlogs, cycle time, citizen satisfaction, procurement. | | sc-edtech (details) | 5 — same | Edtech / online learning. Activation, completion, learning outcomes, monetization. |

All variants produce the same output structure (visual contract). They differ in (a) depth of analysis, (b) industry vocabulary, and (c) common root-cause priors.

Note: the industry packs (sc-saas, sc-retail, sc-healthcare) live in industry-packs/, not skills/. The Quick-start cp -r skills/* does not install them — see industry-packs/README.md for one-line install commands.

The five frameworks

| # | Framework | What it does | Visual format | |---|---|---|---| | 1 | MECE | Categorize factors with no overlap, no gaps | Nested bullet points | | 2 | Issue Trees | Decompose the problem into root causes | ASCII tree | | 3 | Hypothesis-Driven | State a falsifiable hypothesis, design the test | Markdown table (Expected vs. Actual) | | 4 | Pareto Principle | Isolate the 20% of inputs driving 80% of results | Blockquote + deprioritized list | | 5 | "So What?" Test | Translate findings into an actionable insight | Process / Result / Insight |

Each framework is also a standalone skill — invoke /mece-framework on its own when that's all you need. The 7-framework variant adds 5 Whys (drilled root cause) and Pyramid Principle (executive-ready communication) as additional standalone sub-skills.

Who is this for?

Founders & operators prepping for board meetings or investor updates who need to think like a consultant without hiring one.
PMs and analysts structuring messy investigations (churn spikes, conversion drops, support backlogs) before they become slide decks.
Strategy & consulting professionals using it as scaffolding for client deliverables — apply the frameworks fast, then layer in real research.
Engineering managers and TPMs running incident retros or capacity reviews where loose thinking is the enemy.
MBA students & interview candidates — mastering these five frameworks is literally the case-interview job.
Customer success leaders doing churn analysis or QBR prep.
Sales leaders running pipeline reviews — the hypothesis test is especially good for "why are deals stalling?"

Unifying thread: anyone who needs structured business thinking on demand and doesn't want every problem to dissolve into prose.

Design choices

A few decisions worth surfacing — in case you're forking and wondering why things are the way they are:

Visual structure is non-negotiable. Every framework has a strict output format (nested bullets, ASCII tree, table, blockquote, three labeled sections). Prose-only answers are a failure mode. This is the entire value of the skill.
Five frameworks, applied in order. MECE → Issue Tree → Hypothesis → Pareto → So What is the canonical sequence. Skipping or reordering breaks the logic chain. The 7-framework variant adds 5 Whys (between Issue Tree and Hypothesis) and Pyramid Principle (after So What) — also applied in strict order.
Defaults are flexible, MECE is not. Sub-skills say "3–6 categories" or "4–7 variables" — but tell the model to flex with judgment. The one rule that doesn't flex: categories must be Mutually Exclusive and Collectively Exhaustive. That's the foundation; everything else is style.
Three masters, not one. Sequential works everywhere. Sub-agent works only in agent-capable environments. The 7-framework variant is for high-stakes work. Shipping all three lets users pick without losing reach.
Dual format (skills + plain prompts). Anthropic Skills format works in Claude Code/Cowork. Plain prompts work in everything else. Same content, two distributions.
Reframe-the-question is built in. Sometimes the user's stated problem isn't the real problem. The masters ask one reframe question max before applying the frameworks — pushing back is part of the consultant role.

Roadmap

Shipped in v1.x (initial release):

[x] 5-framework master (sequential + sub-agent) ✅
[x] 7 sub-skills, each invokable standalone (5 canonical + 5 Whys + Pyramid Principle) ✅
[x] 7-framework variant — adds 5 Whys (between Issue Tree and Hypothesis) and Pyramid Principle (after So What?) for high-stakes work. See skills/strategy-consultant-7/. ✅
[x] Industry packs — sc-saas, sc-retail, sc-healthcare with industry-specific MECE defaults, vocabulary, and root-cause priors. ✅
[x] Eval suite — automated structural scorer + content rubric in evals/. Run python evals/score.py < output.txt. ✅
[x] Hero illustration at the top of the README. ✅

Shipped in v1.1 (skill + eval expansion):

[x] Industry-pack paste prompts — prompts/sc-saas.md, prompts/sc-retail.md, prompts/sc-healthcare.md (LLM-agnostic versions of each pack). ✅
[x] strategy-consultant-7 fully self-contained — inlines all seven framework specs; works with no sub-skills installed. ✅
[x] Eval suite: 7-framework scoring + fixtures + pytest harness — evals/score.py --mode 7 and evals/test_scorer.py. ✅
[x] Inter-framework data flow — each section feeds the next (MECE → Tree → Hypothesis → Pareto → So What); propagated to industry packs too. ✅
[x] Sub-skill formatting standardization — consistent skeleton across the seven sub-skills. ✅
[x] Follow-up & multi-problem handling in the masters (focused-section follow-ups; multiple-problem disambiguation). ✅
[x] CI workflow — GitHub Action running the scorer + link-checker on every push/PR. ✅

Shipped in v1.2 (latest additions):

[x] Slack-formatted output mode — request "Slack mode" / plain-text (or name Slack/Teams/email as the destination) and the skill renders indented-dash trees instead of box-drawing chars. ✅
[x] LLM-as-judge eval mode — evals/judge.py emits a rubric-based judging prompt for any LLM and grades the JSON reply (content quality 1–5). Dependency-free, no API key. ✅
[x] More industry packs — fintech, manufacturing, nonprofit, government, edtech — shipped as sc-fintech, sc-manufacturing, sc-nonprofit, sc-government, sc-edtech. ✅
[x] More example industries — biotech, gaming, sports, agriculture (added to examples/full-runs/); the repo now ships 25 example files in total. ✅
[x] CI scores real output samples — the pytest harness scores showcase masters/packs (5-mode), the 7-fw showcase (7-mode), and every file in examples/full-runs/ on every PR. ✅

Still in flight:

[ ] Video walkthrough on the ConrayGambit YouTube channel.

If you have ideas, open an issue or DM @ConrayGambit.

Contributing

PRs welcome. The frameworks themselves are well-established consulting practice — the value of this repo is in the strict formatting contracts. If you propose changes, please preserve the visual output structure.

To add a new framework:

Create skills/<your-framework>/SKILL.md with frontmatter
Create prompts/<your-framework>.md as a condensed paste-ready version — output contract only (drop the worked example and meta sections)
Update the master skills to include the new step
Add an entry to the framework table above

To add an example:

Pick an industry not yet represented (see examples/README.md for what's there).
Drop a file into examples/full-runs/ or examples/single-frameworks/ with the input/output structure used by the existing examples.
Update examples/README.md to link to it.

See CONTRIBUTING.md for full guidelines.

License

MIT — see LICENSE. Use freely, modify freely, ship freely.

Credits & Acknowledgment

Acknowledgment: The core frameworks and structural concepts in this skill are directly adapted from the teachings of Analyst Academy on YouTube.

>

Source reference: 5 Consulting Frameworks to Solve Any Problem

The five frameworks themselves (MECE, Issue Trees, Hypothesis-Driven Problem Solving, Pareto Principle, "So What?" test) are classic management consulting practice (McKinsey, BCG, Bain). Analyst Academy's video crystallized the visual output contract — the idea that each framework should produce a specific, repeatable visual artifact (nested bullets, ASCII tree, comparison table, blockquote, three labeled sections). This repo packages that contract as Claude Skills + LLM-agnostic prompts.

The two additional frameworks in the 7-framework variant — 5 Whys (Sakichi Toyoda / Toyota Production System) and Pyramid Principle (Barbara Minto / McKinsey) — are also well-established practice, packaged with the same visual-contract treatment.

About the packager

Built and packaged by Nathaniel Bricknell (@ConrayGambit on GitHub).

For walkthroughs, demos, and more AI/LLM tooling, check out the ConrayGambit YouTube channel — including a video walking through this repo and giving full credit to Analyst Academy for the original framework framing.

If this saved you time

⭐ Star this repo — it helps others find it. 📺 Subscribe to ConrayGambit on YouTube — for more AI/LLM tools, walkthroughs, and prompt-engineering tutorials. 🔗 Share it with the next person you see drowning in unstructured analysis.

🔗 More in this category