Use when designing customer discovery interviews to test problem/customer assumptions, or synthesizing findings from interviews already run, or when 09-lean-canvas.md exists but discovery hasn't started. Step 10 of the idea-to-mvp flow (Lean PMF search).
Two passes, possibly in separate conversations: prep (design the
interviews) and synthesis (make sense of what came back). Detect which
pass applies from what exists in 10-customer-discovery.md and what the PM
brings. If the doc already has a completed ## Synthesis and the PM isn't
asking to redo the script or bringing new notes, there's nothing to do —
say so and point to step 11 (pm-pmf-signal) as the next step. If the PM
brings NEW notes after a completed ## Synthesis (e.g. a second round of
interviews), treat it as another Pass 2 run: re-draft the Synthesis section
from the combined evidence (old + new notes), not just the new batch alone.
Prerequisites
Read 00-status.md and 09-lean-canvas.md. Missing → offer to run
pm-lean-canvas first, or accept an equivalent doc.
Pass 1: Prep (no 10-customer-discovery.md yet, or PM wants to redo the script)
Drafting instructions (dispatch to a subagent when available, else inline)
From the canvas's Problem and Customer Segments boxes, identify the 2-3
riskiest assumptions — the ones that, if false, break the business model.
Draft:
A Mom-Test-style interview script: questions about specific past behavior
and facts ("last time you tried to X, what did you do?"), never opinions
or hypotheticals ("would you use a tool that...?"). 6-10 questions.
Target segment and recruiting criteria (who, and where to find them).
Read each question as if you were the interviewee — does any of them
lead the answer or ask for an opinion instead of a fact?
Is the recruiting criteria narrow enough that everyone you talk to
actually has the problem?
What would you consider a strong enough signal, from how many
interviews, to say "the problem is real"?
Iterate until approved.
Output — save to product/<idea-slug>/10-customer-discovery.md
If ## Synthesis already has real content (a prior synthesis pass ran),
redoing the script means replacing ONLY ## Interview guide — never
overwrite an existing ## Synthesis section back to "pending." A completed
synthesis is downstream input to step 11; discarding it silently is a data
loss, not a fresh start.
# Customer Discovery — <idea title>
## Interview guide
### Riskiest assumptions being tested
### Target segment & recruiting criteria
### Target interview count
### Script
## Synthesis
pending — run pm-customer-discovery again after interviews to synthesize findings