Unpack Problem — Tri-Modal Problem Exploration

A structured problem-exploration workspace. Unlike /validate-idea (which assumes you have an idea), /unpack-problem helps you explore the problem space before committing to any solution. Facilitated by @product-manager and @user-researcher.

When to use

• "Users are dropping off at checkout step 2" — you have a problem, not a solution
• "Our conversion rate is low" — you need to understand why before deciding what to build
• "I know something is wrong but I'm not sure what to fix"
• Problem-first entry — no solution assumption allowed

Primary personas

• @product-manager — owns problem framing, solution mapping, and handoff routing
• @user-researcher — owns user perspective simulation, empathy mapping, and journey analysis

Mode selection

First, detect the user's intent:

• Guided mode → The agent acts as an interactive facilitator, asking probing questions step-by-step. Best for users who want to think deeply through the problem.
• Automated mode → The agent drafts all design thinking artifacts from simulated user perspectives. Best for users who want rapid output to review.
• Combination mode (default) → The agent auto-drafts first, then leads structured review and refinement. Best for most users.

If unclear, present the three modes and ask.

Simulated perspectives are hypotheses, never evidence

Automated/combination modes draft empathy maps, journeys, and JTBD from simulated user perspectives. These are hypotheses to validate with real users — never present them as evidence. Every simulated artifact carries the [AUTO-DRAFT: ...] label (see step-02's dispatch ladder) plus its riskiest assumption, and multi-perspective drafting follows the dispatch ladder in steps/step-02-analysis-execution.md instead of roleplaying all segments in one context.

Halt conditions

• The user proposes a solution before the problem is framed -> halt. Step 01 enforces zero-solution framing; a jumped-to "we should build X" is a solution in search of a problem and must be sent back to the problem statement
• A simulated (auto-drafted) artifact is about to be presented as evidence -> halt. Every tier simulates — native subagents, firewalled calls, and single-context all produce simulated perspectives — so [AUTO-DRAFT: …] is required on all of them, with the reason recorded per tier (see steps/step-02-analysis-execution.md).
• The dispatch ladder is unavailable and multi-perspective drafting would collapse into one shared context -> halt or explicitly mark the output degraded; never silently roleplay all segments in one context
• No solution concept survives the Step 03 mapping -> the brief ends VERDICT: [NO-GO]; do not manufacture a candidate to satisfy the handoff

Done when

• artifacts/output/02-research/problem-space-brief.md exists and ends with a plain sentence plus a bracketed gate line in the document — VERDICT: [RESHAPE] / VERDICT: [NO-GO], or VERDICT: [GO] — When: <NOW|GATED|NEXT-CYCLE> — a brief without a gate is a document, not a decision. Never the unbracketed [GATE: GO|…] spelling, and never in the chat reply.
• The problem statement, root cause, JTBD, and HMW questions are all present and mutually consistent
• All four analysis artifacts (root-cause-analysis.md, empathy-map.md, journey-map.md, jtbd-hmw.md) exist and each matches its shape contract in steps/step-02-analysis-execution.md §Artifact shape contracts
• Every simulated artifact carries its [AUTO-DRAFT: …] label — on every dispatch tier, since all tiers simulate — and names its riskiest assumption; guided-mode artifacts are [USER-SOURCED: …] or labelled
• The handoff routing names the next skill and the artifact it consumes
• @memory-controller session-write has run

Harness adherence (non-negotiable)

• Follow the step sequence for the selected mode exactly. Do NOT skip steps.
• Step 01 enforces zero-solution framing. Do not let the user jump to "we should build X."
• Each step file is a contract. Read it fully before executing.

Step sequence

0. Scope & Decision Anchoring (non-skippable gate) → steps/step-00-scope-and-decision-anchoring.md — step_tracker.js exits 1 if Step 1 begins without a locked track. Intent triage at gate: Clear (C ≥ 0.85) runs the locked track; Ambiguous (0.50–0.85) halts to a 2–3 Track Fork; Trivial (C < 0.50) executes directly. Gate enforced in all step_tracking modes including off (DNA 4).
1. Problem Intake → steps/step-01-problem-intake.md — intake pain points, enforce zero-solution framing
2. Analysis Execution → steps/step-02-analysis-execution.md — run or prompt modular sub-skills based on selected mode
3. Synthesis & Ideation → steps/step-03-synthesis-ideation.md — map problem findings to candidate solution concepts
4. Brief Generation → steps/step-04-brief-generation.md — write problem-space-brief.md (the discovery report is compiled separately by /discovery-report when the user wants a unified report)

Output artifacts

These are the document surface — the machine record lives here, and the gates read these files.

• artifacts/output/02-research/problem-space-brief.md — structured problem definition (template: .agents/templates/discovery/problem-brief.md)
• artifacts/output/02-research/root-cause-analysis.md — root cause findings
• artifacts/output/02-research/empathy-map.md — user empathy canvas
• artifacts/output/02-research/journey-map.md — current-state journey
• artifacts/output/02-research/jtbd-hmw.md — jobs-to-be-done + how-might-we (template: .agents/templates/discovery/jtbd-hmw.md)

Only two of the five ship a template. The other three follow the shape contracts declared in steps/step-02-analysis-execution.md §Artifact shape contracts — that is the rule: an artifact has either a cited template or a declared inline shape, never neither.

The chat reply is a different surface and carries human wording only. State what the problem is and whether it is worth solving, in a sentence, plus the next action — then name the brief. Never put VERDICT:, [GO]/[RESHAPE]/[NO-GO], or any card field in the chat (.agents/references/vespyr-dna.md §Two surfaces).

State machine integration

At start: run node .agents/scripts/orchestrator_state.js status At end: run node .agents/scripts/orchestrator_state.js complete --agent product-manager --artifact 02-research/problem-space-brief.md

Memory integration

At start: Load context before problem intake:

@memory-controller load product-manager [unpack-problem — {problem description}]

At step-04 (brief generation): Write session summary — mandatory:

@memory-controller session-write [agent: @product-manager]
Worked on: Problem space exploration — {problem description}
Decisions: {key problem framing decisions, root causes identified, candidate solutions mapped}
Next step: {validate-idea / shape-up / explore-idea — whichever handoff routing applies}
Blockers: {unresolved problem areas or "none"}

Handoff routing

The brief ends with a Decision Gate on the selected solution concept — plain sentence first, then VERDICT: [RESHAPE] / VERDICT: [NO-GO] (or VERDICT: [GO] — When: <NOW|GATED|NEXT-CYCLE>), never the unbracketed [GATE: GO|…] form (protocol in steps/step-04-brief-generation.md). A brief without a gate is a document, not a decision.

• Solution concept ready to test → load /validate-idea
• Structured brief ready for specs → load /shape-up
• Need market/competitor context → load /explore-idea