Product strategy

Contents

Reference	When to load
references/orchestrator.md	Running the pipeline — five phases plus a critic pass
references/write-protocol.md	Canonical write-protocol block copied verbatim into every subagent prompt
references/frameworks/INDEX.md	Routing guide — decision table + cross-framework composition
references/frameworks/*.md	Per-framework files — Rumelt kernel, Wardley maps, Minto pyramid, Working Backwards
references/roles/cpo.md	CPO role — product-line framer
references/roles/cso.md	CSO role — portfolio framer (build-vs-buy, M&A, 3–10 year horizon)
references/roles/vp-design.md	VP Design role — customer-archetype and experience framer
references/roles/rumelt-architect.md	Runs the three-part kernel plus The Crux
references/roles/wardley-cartographer.md	Builds the text-form Wardley map
references/roles/pr-faq-discovery.md	Drafts a PR-FAQ as a discovery artifact (forcing function for customer-value clarity)
references/roles/minto-pyramid-builder.md	Composes SCQA + MECE argument pyramid
references/roles/strategy-synthesizer.md	Reads every packet, composes the final strategy memo
references/roles/strategy-critic.md	Reviews the memo on one axis (coherent + defensible), multi-dimensional rubric
templates/worklog-skeleton.md	Per-role work-log skeleton with embedded write-protocol
templates/strategy-memo.md	Final deliverable skeleton — Executive Summary, Diagnosis, Guiding Policy, Actions, Risks
templates/rumelt-packet.md	Rumelt architect's output shape
templates/wardley-packet.md	Wardley cartographer's output shape
templates/pr-faq-discovery.md	PR-FAQ skeleton — press release + external FAQ + internal FAQ
templates/minto-outline.md	SCQA + MECE outline

Product strategy

One skill, one pipeline. Takes a strategic question → polished strategy memo through five phases that write their outputs to disk in real time. Framework agents in Phase 2 run in parallel. Synthesis and critic pass happen in sequence. The grouped framework file loads on demand — its decision table routes the orchestrator without every framework body being read.

Pipeline at a glance

flowchart LR
    A[Phase 0<br/>Route] --> B[Phase 1<br/>Frame<br/>CPO / CSO / VP-Design]
    B --> C[Phase 2<br/>Parallel framework agents]
    C --> C1[Rumelt architect]
    C --> C2[Wardley cartographer]
    C --> C3[PR-FAQ discovery]
    C --> C4[Minto pyramid builder]
    C1 --> D[Phase 3<br/>Synthesize<br/>strategy-synthesizer]
    C2 --> D
    C3 --> D
    C4 --> D
    D --> E[Phase 3.5<br/>Critic review<br/>strategy-critic]
    E --> |Revise up to 2 rounds| D
    E --> F[Phase 4<br/>Deliver + handoff]

Every role runs as a general-purpose Agent with model: "opus" and a prompt built from the matching file in references/roles/. Phase 2 fans out in parallel with run_in_background: true. Full runbook with prompts, check-in cadence, and stuck detection lives in references/orchestrator.md.

When to run which role

User signal	Fan-out
"Run a Rumelt kernel" / "what's the crux" / "our strategy feels fluffy"	rumelt-architect only
"Build a Wardley map" / "should we build or buy X" / "platform strategy"	wardley-cartographer only
"Draft a PR-FAQ for a new product idea"	pr-faq-discovery only
"Structure this argument" / "build a Minto pyramid"	minto-pyramid-builder only
"Write an executive strategy memo" / "full strategy read"	all four frameworks, full memo
"Portfolio bet" / "build vs buy across the portfolio"	CSO framing + wardley + rumelt
"Customer-centric new product"	CPO + VP-Design + pr-faq + rumelt
"Review my existing strategy memo"	critic only
Targeted question on one sentence / section	Inline — no subagents

Run end-to-end with no approval gates between phases when the ask is clear. The pipeline is fully reversible — every file is on disk; nothing is published. Surface the plan inline, proceed. Use one AskUserQuestion only when audience, horizon, or deliverable shape is genuinely ambiguous.

Write-protocol discipline

Every role follows the same rhythm: one unit of thought → edit the output file → next unit. Partial work on disk survives timeouts and context pressure; state held in working memory does not. The canonical block lives in references/write-protocol.md and is copied verbatim into every role prompt and every task skeleton — one source of truth, no paraphrasing.

Output files in product-strategy/{{ slug }}/ are the source of truth: framing.md, rumelt-packet.md, wardley-packet.md, pr-faq-packet.md, minto-outline.md, strategy-memo.md, review-strategy.md. Nothing else is load-bearing.

Framework family

Four frameworks ship as one routing guide plus four per-framework files. The INDEX's decision table routes the fan-out; per-framework files load only when a role needs them.

• Rumelt Kernel + The Crux — diagnosis, guiding policy, coherent actions; the Crux names the single pivotal challenge. Use for: roadmap prioritization, "our strategy is fluff," portfolio bets with unclear diagnosis.
• Wardley Maps — user need + value chain + evolution axis (Genesis → Custom → Product → Commodity) + climatic patterns + gameplay. Use for: build-vs-buy, platform strategy, open-source decisions.
• Minto Pyramid — SCQA + MECE-grouped supporting arguments. Thinking structure for the memo; also runs standalone when the ask is "structure this argument."
• Working Backwards PR-FAQ — press release + five customer questions + internal FAQ. Used here as a discovery artifact (forcing function for customer-value clarity), not a publication-ready document.

Full decision table and cross-framework composition guidance in references/frameworks/INDEX.md. Per-framework canonical structure, When-to-use / When-to-skip, templates, and citations live in one file per framework under references/frameworks/ — rumelt-kernel.md, wardley-maps.md, minto-pyramid.md, working-backwards.md.

Narrative framework composition

• Strategy feeds discovery. strategy-memo.md is the upstream input to product-discovery's PRD derivation — HMW questions and job stories derive from the memo's customer framing and coherent actions.
• Strategy feeds long-form documents. strategy-memo.md + pr-faq-packet.md carry the Minto-shaped thinking; hand them off to whatever narrative-writing workflow your team uses.

The framework reference's "Cross-family composition" section names the common lineages — Working Backwards often follows a Rumelt kernel; Minto structures the output of either; Wardley feeds diagnosis when the challenge depends on "what's commoditizing."

When NOT to use this skill

• Writing a PRD, HMW questions, job stories, EARS specs, user stories. Use product-discovery.
• Writing a long-form narrative. Strategy frameworks produce the thinking — hand the memo and PR-FAQ to whatever narrative-writing workflow your team uses.
• Pure execution work. If the diagnosis is already agreed and the team is building, strategy frameworks add overhead without insight.
• Tactical decisions. "Which color for the button" does not earn a kernel.

Anti-patterns

• Don't fan out all four frameworks by default. The Phase 0 routing picks the subset that matches the ask — over-fanning wastes Opus compute and dilutes the synthesis.
• Don't skip the framing phase. Without framing.md, the Phase 2 roles don't know what question they are answering, and the synthesizer merges artifacts that don't share a thesis.
• Don't let the synthesizer re-run a framework. If a packet is thin, relaunch the framework role with the thin packet preloaded — don't paper over it in the memo.
• Don't run more than 2 review-revise iterations. After 2 rounds, surface remaining findings inline and hand the memo to the user rather than looping.
• Don't reproduce framework bodies in the memo. The memo uses the output of each packet (diagnosis, map-read, customer archetype, MECE groups), not the framework's canonical structure.
• Don't drop a strategy memo on an execution problem. If the real ask is "write a PRD," route to product-discovery instead of forcing a kernel.