Customer discovery

Customer Discovery (experimental — "Framework plus")

Run synthetic customer discovery for one opportunity and produce a preliminary draft of Worksheet 2's Factor 1 — Compelling reason to buy, from grounded, deliberately skeptical synthetic personas.

Experimental "Framework plus" — read this first. Synthetic personas are, by default, sycophantic and ungrounded: an LLM will happily tell you your idea is great. This skill fights that (web-grounded personas, a skeptic/loyalist/laggard spread, explicit anti-agreeableness instructions), but it still produces a PRELIMINARY DRAFT to break the blank-page problem and to prep real interviews — it is not a substitute for getting out of the building. Real customer contact stays the gold standard. Keep the human in the loop.

Per the project owner's explicit decision, this skill does output a draft score for Compelling reason to buy (Low / Mid / High / Super High) that the team can accept or override — alongside the load-bearing assumptions and an interview kit for the real conversations.

How to run it — detect the engine first

Step 0 — capability check. Before anything else, check whether a Workflow tool is available on this surface. That one fact picks the engine: IF Workflow is available (Claude Code) → run the bundled workflow (Path A). IF it is NOT (Cowork / Agent-SDK and other surfaces) → run the in-skill fallback (Path B). Never block on the missing tool.

Path A — Workflow (Claude Code)

Call the bundled workflow (ships with the plugin — no install step):

Workflow({
  scriptPath: "${CLAUDE_PLUGIN_ROOT}/workflows/wf-customer-discovery.js",
  args: { opportunity: "<name/description>", value_prop: "<the compelling reason claimed>", workspace: "where-to-play" }
})

You may also pass application, customer and factor. The workflow: (1) one grounded agent derives 4–6 diverse personas via web search — separating the economic buyer from the user, and including a skeptic, an incumbent-loyalist and a laggard, not only the eager early adopter; (2) one agent per persona role-plays a realistic, non-sycophantic interview (real objections, status-quo inertia, "I already have a solution", budget reality); (3) one synthesis agent returns a draft score + honest red/orange/green confidence + rationale + key assumptions + a Mom-Test interview kit + a SYNTHETIC / directional-only note.

Path B — in-skill fallback (Cowork / Agent-SDK)

If Workflow is unavailable, degrade gracefully and reproduce the same flow yourself. (The customer personas are derived dynamically per opportunity — there are no bundled customer-persona agents to call, and the persona-explorer / persona-skeptic agents are idea/scoring-debate lenses, not customers — so generate the personas here rather than spawning fixed agent ids.)

1. Generate 4–6 diverse, skeptical personas (buyer vs. user; include a skeptic, an incumbent-loyalist and a laggard), grounded with web search where you have it.
2. Interview each persona — spawn them as subagents where supported, otherwise role-play each sequentially in-skill. Instruct each to resist agreeableness and voice real objections, switching barriers and budget reality.
3. Synthesise the same output: a draft score (Low/Mid/High/Super High), honest red/orange/green confidence, rationale, key assumptions to validate, and a 5–8 question interview kit.

Note: in-skill role-play shares one context across personas (less isolation than separate subagents), so the personas are less independent — treat the fallback as lower rigor.

After it reports

1. Present the draft to the team as a starting point, not a verdict: the proposed Compelling reason to buy score (Low / Mid / High / Super High) with its red / orange / green confidence, the rationale, the diverging buyer-vs-user signals, and the switching barriers.
2. Write it into the workspace. Put the draft score + confidence into the Factor 1 section of where-to-play/02_attractiveness-map/<opportunity>.md (or a customer-discovery note beside it), clearly labelled SYNTHETIC / DRAFT. Add the key assumptions to where-to-play/assumptions.md.
3. Feed the revisit loop: draft now → run the interview kit with real customers → re-score. Route the (re)scoring itself to the attractiveness-map skill — the team sets the final score, never this skill.

Boundaries

• Experimental "Framework plus" — the draft score is provisional and synthetic, a prompt to learn, not a decision. The team accepts or overrides.
• Never let synthetic feedback stand in for real customer contact; the interview kit exists precisely to get the team out of the building.
• Canonical terminology only: Compelling reason to buy; scores Low / Mid / High / Super High; confidence red / orange / green.