Pm

PM Router

Turn a product request into a named decision, route it to one primary skill and at most two companions, then produce a decision-ready artifact.

Route

1. Restate the product, customer, business objective, constraint, and decision in one compact block. Mark missing facts as unknown; do not invent them.
2. Identify the dominant uncertainty: problem, demand, value, usability, feasibility, viability, evidence, alignment, or AI behavior.
3. Select the primary skill from the table. Add a companion only when its output is a prerequisite or guardrail.
4. Load the selected skills. In Claude, invoke them by name. In Codex, read their installed SKILL.md files.
5. Execute the leaf workflows. Do not merely describe their frameworks.
6. End with the decision, evidence, unresolved risk, owner, and next learning action.

Routing table

User intent or decision	Primary skill	Common companion
Find or validate customer problems	product-discovery	product-interview
Plan or analyze customer conversations	product-interview	product-discovery
Map outcomes, opportunities, solutions, and tests	product-opportunity-tree	product-discovery
Choose focus, trade-offs, or coherent direction	product-strategy	product-prioritization
Assess target customer, unmet need, or fit	product-market-fit	product-metrics
Establish competitive context and value	product-positioning	product-launch
Choose price, package, or value metric	product-pricing	product-positioning
Define success, metric trees, or instrumentation	product-metrics	product-experiment
Design a test that changes a decision	product-experiment	product-metrics
Rank bets under constraints	product-prioritization	product-strategy
Turn strategy into an outcome roadmap	product-roadmap	product-strategy
Write a lean product brief, PRD, or PR/FAQ	product-prd	product-metrics
Prepare rollout, adoption, or launch gates	product-launch	product-positioning
Align stakeholders or communicate a decision	product-stakeholders	product-strategy
Decide whether AI is appropriate	product-ai-fit	product-ai-risk
Define AI quality and release thresholds	product-ai-evals	product-ai-risk
Analyze AI failure, safety, reliability, or monitoring	product-ai-risk	product-ai-evals
Review a product bet or evidence packet	product-review	Route to the weakest decision area

Companion rules

• Discovery plus interview: use when evidence must be collected, not merely synthesized.
• Strategy plus prioritization: strategy sets the rule; prioritization applies it. Never reverse that dependency.
• Metrics plus experiment: metrics define the observable change; the experiment defines the causal learning plan.
• Positioning plus launch: positioning establishes context before launch messaging or channel choices.
• AI fit plus risk plus evals: use all three only for a consequential AI launch decision.
• For UI design, keep /design responsible for interface execution and use /pm for customer, business, evidence, and product-decision framing.

Shared evidence standard

Label important claims as:

• Observed: behavior, telemetry, transaction, or directly inspected artifact.
• Reported: a participant or stakeholder statement.
• Inferred: a conclusion drawn from evidence.
• Assumed: unverified input required to proceed.

Prefer observed behavior and commitments over opinions, compliments, and hypothetical intent. Match evidence strength to reversibility and downside.

PM quality gate

Before answering, verify:

• A real decision and decision owner are named.
• The target customer and desired outcome are specific.
• Customer value, usability, feasibility, and business viability are considered.
• Evidence is separated from assumption.
• Alternatives and explicit non-goals are visible.
• Success and guardrail measures are decision-linked.
• The next action reduces the largest uncertainty at proportionate cost.
• A review date, trigger, or stop rule prevents indefinite drift.

Escalate to a responsible human for legal, privacy, safety, employment, regulated-domain, irreversible pricing, or high-impact automated-decision approval.