decompose

name

decompose

description

Pick the right consulting/strategy framework to structure a messy problem. Routes a problem statement through a 5-axis router (purpose, horizon, locus, unit, knowability) over an atlas of ~80 frameworks — McKinsey, BCG, Porter, root-cause, prioritization, systems, change, risk, process, negotiation, and more. YOU GET BACK: the problem's shape reflected in one line, 2-3 candidate frameworks (recommended marked ●, alternatives ○) with one-line fit reasons, a proposed composition when one lens won't carry it (often 2-3 in sequence), and the recommended framework's failure mode flagged for your specific case — then an offer to actually apply it. Fires when the user has a tangled, ambiguous, or open-ended problem and wants structure: 'how should I think about X', 'what framework fits this', 'help me break this down', 'decompose this', 'what would a consultant do here'. Also fires on '/decompose', 'framework?', or naming a framework family ('Porter', 'BCG matrix', 'issue tree', 'Eisenhower'). Most valuable mid-problem-statement, before the user commits to a shape. Do NOT fire for execution tasks ('write the code', 'draft the email') or once a framework is chosen and the user is executing.

decompose

Pick the right lens. Compose two or three if no single one covers the problem.

The atlas (atlas.md next to this file) lists ~80 frameworks across 14 families. Don't dump it on the user. Route to a small set, justify the pick, and propose a composition.

When this fires

The user has a problem with no clear shape. They're trying to think clearly about it. Maybe they've named a framework already (good — anchor there, then check if a better fit exists). Maybe they haven't (better — route from purpose).

The router

Classify the problem along five axes. Don't ask all five — infer from the user's framing, then ask only the missing discriminators.

Axis	Poles	Discriminates
Purpose	diagnose / decide / communicate / execute	Issue Tree vs RICE vs Pyramid vs Kotter
Horizon	now / quarter / 1-3yr / 5yr+	Eisenhower vs OKR vs Ansoff vs Scenarios
Locus	internal / external / interface	7S vs PESTLE vs Five Forces
Unit	task / product / unit / portfolio / industry	MoSCoW vs JTBD vs BMC vs BCG Matrix vs Porter
Knowability (Cynefin)	clear / complicated / complex / chaotic	SOP vs Six Sigma vs Scenarios/Wardley vs OODA

If purpose is unclear, ask first. It does the most discriminating work.

Purpose → framework families

Once purpose is named, the candidate set narrows fast:

• Diagnose cause → Issue Tree, 5 Whys, Fishbone, Iceberg, Causal Loop, Fault Tree
• Diagnose position → SWOT, Five Forces, 7S, Wardley, GE 9-box
• Prioritize options → RICE / ICE, MoSCoW, Eisenhower, WSJF, Weighted Scoring, Kano
• Allocate across portfolio → BCG Matrix, GE 9-box, Three Horizons, Three Box, Real Options
• Choose growth path → Ansoff, JTBD, Blue Ocean (ERRC), Wardley, Disruption
• Plan change → Kotter, ADKAR, 7S, Influence Model, Lencioni, Tuckman
• Manage risk / uncertainty → Pre-mortem, Scenarios, Cynefin, Real Options, Bayesian
• Improve process → Lean (7 Wastes), DMAIC, Theory of Constraints, VSM, SIPOC, A3
• Communicate decision → Pyramid Principle, Custom 2×2, Strategy Canvas
• Understand customer → JTBD, STP, Kano, Customer Journey, AARRR
• Negotiate → BATNA/ZOPA, Game Theory
• Learn faster → OODA, Double-Loop, PDCA

How to respond

Three sentences of routing, then the recommendation. Don't lecture.

1. Reflect the problem shape back in one sentence — purpose, horizon, unit, knowability. Show the user you heard it.
2. Name 2-3 candidate frameworks with one-line reasons. Mark the recommended one with ●, alternatives with ○.
3. Propose a composition if one framework won't carry it. Real problems usually need a sequence.
4. Flag the failure mode of the recommended framework in the user's specific situation — if the canonical failure matches their context, swap or supplement.

Then offer to actually apply it — fill in the issue tree, populate the 2×2, run the pre-mortem, whatever the chosen framework asks for.

Applying the pick

When the user says "do it," choose the execution shape BEFORE touching the frameworks. Depth scales with independence, not context length — one context applying four lenses gives each lens a quarter of a head.

Shape	Condition	Execution
Inline	One framework; or a dependent chain where each output feeds the next (SWOT → TOWS); or placement/scoring a single pass can do (force-placing 9 clients on a 2×2)	Apply serially, here
Fan-out	≥3 independent applications, each substantial enough to want its own agent's full attention — parallel lenses on one problem, deep branches of an issue tree, quant + qual instruments built independently	Author a Workflow via the workflow-author skill

Burden of proof is on orchestration. Fanning out agents to score items one pass could place is theater — one subagent is not a workflow.

For fan-out, invoke workflow-author and hold its contract:

• One agent per framework/lens/branch, each given the problem statement + the atlas row for its framework (shape, inputs, failure mode) — agents don't get the whole atlas.
• Force structured output with schema: findings as claims + evidence + confidence + the framework's failure-mode check applied to this case. Agents return raw findings, not prose.
• pipeline by default; a barrier (parallel) only where synthesis genuinely needs every result at once — cross-framework contradiction hunting, quant-vs-qual comparison.
• Mixed compositions split: fan out the independent stages, keep the dependent spine serial (PESTLE ∥ Five Forces ∥ 7S in parallel → SWOT/TOWS synthesis inline).
• Synthesize in the main context. The weave — where lenses agree, where they contradict, what that means — is the deliverable. Never paste raw agent output as the answer.

Composition patterns (worth knowing cold)

• Strategy refresh: PESTLE → Five Forces → SWOT → TOWS → Ansoff → OKRs
• Product prioritization: JTBD → Kano → RICE → Pyramid
• Org transformation: 7S diagnostic → Influence Model → Kotter → ADKAR tracking
• Ops problem: SIPOC → VSM → 5 Whys / Fishbone → DMAIC
• New venture: Lean Canvas → JTBD → Wardley → Blue Ocean → AARRR
• High-stakes decision under uncertainty: Cynefin classify → Scenarios → Pre-mortem → Real Options framing
• Personal triage: Eisenhower → 80/20 → OKR alignment check

Anti-patterns

• Picking a framework before stating the question. The framework should fall out of the problem.
• Using a 2×2 when the real shape is a tree (hides branches).
• Using a tree when the real shape is a loop (misses feedback).
• Treating qualitative frameworks (SWOT, 7S) as analysis output rather than scaffolding for further work.
• Stacking 5+ frameworks per problem — diminishing returns past 2-3.
• Reaching for the prestigious framework (Porter, BCG) when a humble one (Eisenhower, 5 Whys) is exactly right.

The atlas

Full inventory is in atlas.md adjacent to this skill. Reach for it when:

• The router lands you in a family but you need to compare specific frameworks within it.
• The user names a framework you want to confirm details on.
• You want to surface the failure mode of a candidate.

Don't paste the atlas into the chat. Quote the relevant rows.