management-consulting — Strategy Skill for AI Coding Agents

Turn your AI agent into a structured thinking partner for strategy, decisions, and analysis. 42 frameworks with practitioner-grade depth — not textbook theory, but step-by-step workflows, real examples, common mistakes, and when each framework breaks down.

Built for Claude Code. Compatible with Codex, Gemini CLI, Cursor, and any agent that reads SKILL.md files.

Why this exists

Asking an LLM to “act like a consultant” produces generic framework-salad. This skill fixes that by providing:

• Process discipline — clarify the decision before analyzing. Structure before hypothesizing. Evidence-label every claim.
• Framework depth — 4,100+ lines of practitioner workflows across 42 frameworks. Each has step-by-step application, good/bad output examples, common mistakes, and failure conditions.
• Three modes — Quick Structure (exploratory), Full Case (analytical), Client Deliverable (polished). Matches response depth to your actual need.
• Quality gates — output contract, devil’s advocate step, 9-point checklist. Prevents shallow answers.
• Anti-selection rules — tells the agent when NOT to use a framework. Stops Five Forces on every strategy question.

Install

# Claude Code
cp -r management-consulting/ ~/.claude/skills/management-consulting/

# Or use npx
npx skills add gcamilo/management-consulting

Usage

Just ask naturally. The skill activates on strategy questions, decisions, market analysis, due diligence, or structured recommendations.

Quick Structure — thinking out loud, exploratory

“Help me think through build vs buy for our data pipeline”

Full Case — specific decision with real stakes

“Should we expand into the mid-market segment?”

Client Deliverable — deck-ready, stakeholder-facing

“Draft a strategy memo for the board on our market entry options”

What’s inside

management-consulting/
├── SKILL.md (258 lines)                    # Core: modes, workflow, output contract
├── references/
│   ├── frameworks-strategy.md (798 lines)  # Five Forces, Value Chain, 7S, VRIO, Blue Ocean...
│   ├── frameworks-financial.md (781 lines) # Profit Tree, Revenue Tree, TAM/SAM/SOM, Unit Econ
│   ├── frameworks-decision.md (708 lines)  # RAPID, Decision Matrix, Pre-Mortem, Inversion
│   ├── frameworks-pricing.md (688 lines)   # Value-Based, Cost-Plus, Competitive, Dynamic, Freemium
│   ├── frameworks-problem-solving.md (511) # MECE, Issue Tree, Hypothesis Tree, Pyramid
│   ├── frameworks-operations.md (484)      # ADKAR, Kotter, Lewin, Customer Journey, NPS
│   ├── frameworks-innovation.md (390)      # JTBD, Business Model Canvas, Advantage Matrix
│   ├── diagram-templates.md (231 lines)    # 12 diagram types (SVG templates)
│   ├── tooling-appendix.md (69 lines)      # Optional: rendering pipeline, themes
│   ├── frameworks-index.md (59 lines)      # Quick-reference: all 42 frameworks, one-line each
│   └── evidence-standards.md (24 lines)    # Claim labeling: Fact / Inference / Assumption / Estimate
├── examples/
│   ├── end-to-end-full-case.md (117)       # SaaS mid-market expansion (complete walkthrough)
│   ├── issue-tree-good-vs-bad.md (78)      # Before/after comparison
│   ├── recommendation-good-vs-bad.md (45)  # Before/after comparison
│   └── end-to-end-quick-structure.md (36)  # Build vs Buy (light mode)
└── variants/
    ├── strategy-memo.md (102)              # Pyramid Principle + SCQA + deck outline
    ├── due-diligence.md (81)               # Commercial DD + red flags
    ├── market-entry.md (80)                # Five Forces + TAM + positioning
    └── org-effectiveness.md (62)           # 7S + ADKAR + stakeholder mapping

22 files. ~5,700 lines. ~333K chars.

Framework depth

Every framework reference file includes (per framework):

42 frameworks across 7 categories

Visual Reference Guide — interactive cards with diagrams, examples, step-by-step mechanics, and key questions for each framework.

Strategy (11): Five Forces, Value Chain, 7S, VRIO, 3 Horizons, Ansoff, Growth-Share Matrix, Nine-Box, Blue Ocean, Wardley Mapping, Scenario Planning

Problem Solving (6): MECE, Issue Trees, Hypothesis Trees, 7-Step Problem Solving, Pyramid Principle, SCQA

Decision Making (5): RAPID, Decision Matrix, Pre-Mortem, Second-Order Thinking, Inversion

Financial (5): Profit Tree, Revenue Tree, TAM/SAM/SOM, Unit Economics, Waterfall/Bridge

Operations (6): ADKAR, Kotter’s 8-Step, Lewin’s Change Model, Influence Model, Customer Journey, NPS

Innovation (4): Jobs to Be Done, Business Model Canvas, Value Proposition Canvas, Advantage Matrix

Pricing (5): Value-Based, Cost-Plus, Competitive, Dynamic, Freemium

How it was built

1. 3-way AI research — Gemini, Codex, and Claude independently researched all 42 frameworks in parallel (21 research tasks, 497K chars of raw material)
2. Synthesis — 4 agents merged the best from each AI per category. Took the sharpest examples, cut textbook filler, resolved disagreements.
3. 3 rounds of adversarial review — Codex (scored 7/10), Gemini (scored 9.4/10), and self-review identified gaps in portability, completeness, and actionability. All addressed.
4. A/B eval — 7-way comparison of different reference configurations to find the highest-impact bundle.
5. v2 eval redesign — 9-criteria rubric with programmatic checkers, pairwise comparison with position swap, and Gemini+Codex adversarial review of the eval plan itself. Key finding: baseline beating the skill on some prompts revealed a skill bug (hard gate too aggressive — the skill stopped at clarifying questions on vague prompts instead of stating assumptions and proceeding), not a measurement problem. But the rubric redesign was needed to properly measure process discipline, which the v1 absolute scoring missed.

Eval results

Latest: Post-polish (Apr 29 2026) — skill-eval v2 rubric

Re-run after the Apr 29 polish (mode-detection cues, KILL CONDITIONS section, [E] evidence label, context-check rename). Gemini-flash judge, Sonnet generation, skill_only approach (no examples bundle).

Skill wins 3/3 prompts:

By tier:

The hard prompts show the polished skill correctly applying Quick Structure mode — shorter, focused responses that ask for the data needed before going deeper. On the framework-salad adversarial prompt, Decision Quality scored 10.0 (judge agreed that pushing back on the 4-framework request was the right call), even though Process Discipline scored lower because the response was short. The rubric still rewards heavy structure; the skill now correctly trades structure for responsiveness on vague/adversarial inputs.

Pre-polish baseline (Apr 24 2026)

Anthropic skill-creator blind A/B (/skill-creator, Opus judge)

The judge identified process discipline as the key differentiator — the skill enforces structured framing and evidence labeling that the baseline skips. Not yet re-run after the Apr 29 polish.

Custom dual-judge eval (skill-eval v2, Gemini 3.1 Pro + Claude Sonnet)

Notable findings:

• Easy prompts show the largest T1 gap (+4.8) — the skill enforces structure that the model naturally skips on simple questions
• Evidence labeling is the sharpest signal — 48–99 labeled claims per response vs 0 baseline; no overlap between distributions

Customization

• Add variants — variants/your-scenario.md with upfront questions, required frameworks, deliverables
• Add examples — examples/your-example.md showing input → mode → output
• Adjust strictness — edit hard gate, output contract, or quality checklist in SKILL.md
• Add frameworks — create or extend references/frameworks-{category}.md

License

MIT