mckinsey-strategy-team

A Claude Code skill that orchestrates a live agent team for strategic decisions. A team lead runs intake, classifies the question, then runs the team in waves on McKinsey-style frameworks: diagnose + map the market in parallel, generate competing options from opposing mandates and filter them to a shortlist, synthesize a recommendation, then turn a panel of verifiers loose on its load-bearing assumptions (a weakness ≥2 of them agree on doesn’t survive) before it ships. Output: a board-ready decision memo + narrative that survives a leadership or board meeting.

Use it to: prepare a leadership/board session, structure a merger/M&A question, or stress-test / war-game an existing strategy.

What’s inside

mckinsey-strategy-team/
├── SKILL.md        # the team-lead recipe (the orchestration)
├── README.md       # this file
├── WHY.txt         # the design rationale (what, how, value)
└── references/     # 21 McKinsey-style frameworks, 6 domains (method docs)

The 21 frameworks come from github.com/aapersh/strategy-skills-for-claude. They are not installed as separate skills — the teammates read them as files. That sidesteps the limitation that a teammate does not inherit a subagent’s skill frontmatter.

Credits

The framework reference docs in references/ are based on Oria AI’s “21 Strategy Skills for Claude” — oria.one/resources/21-strategy-skills-for-claude. This project adapts that material into a live agent-team orchestration; all the underlying McKinsey-style methodology and framework write-ups originate from their work. Huge thanks to the Oria team for making it available — bedankt! 🙏

Prerequisites

• Claude Code ≥ 2.1.32 (claude --version)
• Agent teams enabled. Quickest way — run this in your terminal before you start:

  export CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
  

  That only applies to the Claude Code session you launch from that shell. To enable it everywhere (including the desktop/web app), set it in ~/.claude/settings.json instead:

  { "env": { "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1" } }
  

• Works in any terminal — the default "teammateMode": "in-process" is all you need.

Install

Option A — user skill (available everywhere):

# put this folder wherever you like, then symlink it:
ln -s "$(pwd)/mckinsey-strategy-team" ~/.claude/skills/mckinsey-strategy-team

Option B — per project:

cp -R mckinsey-strategy-team .claude/skills/mckinsey-strategy-team

At runtime the lead resolves its references/ path, trying the user-level install (~/.claude/skills/mckinsey-strategy-team/references) first and falling back to a per-project install ($PWD/.claude/skills/mckinsey-strategy-team/references) — so either install option above works. The only requirement is that the install folder keeps the name mckinsey-strategy-team.

Use

Start a new Claude Code session and say, for example:

“Use the strategy team to prepare a leadership decision on whether we do [X] or [Y]. Here’s the context: …”

The lead runs a short intake and shows you an engagement plan (the wave plan, which frameworks, standard vs deep mode) before any teammates are spawned. Then the pipeline runs: classify → diagnose + market → generate-and-filter options → synthesize → verifier-panel pressure-test → decision memo. Deep mode adds a tournament to rank surviving options and a loop that keeps hunting risks until the panel runs dry.

A standard run is ~7 live sessions across its waves (more in deep mode) — noticeably more tokens than an ordinary chat. Use it for real decisions, not quick questions.

Not for this

• Applying a single framework on its own → just load that framework directly.
• A quick question that doesn’t justify a multi-agent run → use a single session.

License

MIT — see LICENSE. This covers the orchestration and docs in this repo. The framework reference material is adapted from Oria AI (see Credits); please keep that attribution when reusing it.