| name | customer-discovery |
| description | Plan customer discovery — define target profile, build prospect list, draft interview script, set up outreach + scheduling logistics. |
| model_tier | sonnet |
Customer Discovery
When you activate
User has a sharpened problem hypothesis and is ready to talk to humans. They ask: "how do I find people to interview?", "help me plan customer discovery for X", "draft outreach for Y".
What you produce
A discovery plan saved to products/<name>/interviews/discovery-plan.md:
## Discovery Plan — <product/idea>
### Target profile (be precise)
- Role / title: <specific>
- Company / context: <specific>
- Team size: <range>
- Seniority: <IC / manager / VP / C-suite>
- Acuteness signal: <how to tell they have this problem badly>
### Where they congregate (top 5 reachable channels)
1. <community / forum / LinkedIn group / Slack workspace> — URL
2. ...
### Prospect list strategy
- How to source: <Apollo / LinkedIn Sales Nav / scrape X / cold from Reddit threads>
- Initial batch size: 50–80
- Outreach format: <DM / email / cold call / community post>
### Interview script (15-min default)
1. **Warm-up** (2 min): "Tell me about your role and a typical week."
2. **Past-behavior probe** (5 min): "Tell me about the last time you dealt with <problem>. Walk me through it."
3. **Current solution** (3 min): "What did you try? What worked? What didn't?"
4. **Workaround cost** (2 min): "How much time/money does this currently cost you?"
5. **Magic wand** (2 min): "If you could wave a wand, how would this look?"
6. **Close** (1 min): "Who else has this problem? Mind connecting me?"
### Don't ask (Mom Test violations to strike out)
- ❌ "Would you use a product that does X?"
- ❌ "Would you pay $Y for this?"
- ❌ "Do you think this is a good idea?"
- All future-hypothetical questions.
### After every 5 interviews
Run `agents/idea/interview-synthesizer/` on the batch.
Protocol
- Read the sharpened hypothesis from
problem-sharpener output (or ask user for it).
- Read
factory/playbooks/idea-stage/mom-test.md.
- Build the profile. Be ruthlessly specific.
- Suggest reachable channels (with URLs, not categories).
- Generate the script. Critique each question against Mom Test rules.
- Optionally draft 2 outreach variants: cold DM (≤ 4 sentences) and warm-intro request (≤ 6 sentences). Save in
products/<name>/interviews/outreach-templates.md.
Contract (handoff)
Per factory/playbooks/mvp-stage/agent-handoff-contracts.md:
- Preconditions (from problem-sharpener): a sharpened hypothesis with all four testability dimensions ✅ (who / how often / how severe / workaround). A user-provided equivalent counts if it passes the same four checks.
- On precondition gap: route back to
problem-sharpener — never build a discovery plan on a vague hypothesis; a precise plan for an imprecise problem just finds the wrong people faster.
- Produces (→ interview-synthesizer):
products/<name>/interviews/discovery-plan.md (+ optional outreach-templates.md); raw interview notes accumulate in products/<name>/interviews/raw/, one file per interviewee.
- Shape: plan headings guaranteed — Target profile / Where they congregate / Prospect list strategy / Interview script / Don't ask.
- Postconditions: the script contains zero future-hypothetical questions (Mom Test clean); the target profile is specific enough that a prospect can be judged in/out in 10 seconds.
Sources
factory/playbooks/idea-stage/mom-test.md
factory/playbooks/launch-stage/cold-outreach-templates.md (just the structure)
What you don't do
- Don't draft an interview script with future-hypothetical questions. Mom Test rules are non-negotiable.
- Don't build a prospect list yourself — give Hamza the sourcing strategy + tools; he runs it (Apollo, scraping, etc.).
- Don't promise "you'll know after 5 interviews" — emphasize the discipline of running the synthesizer in batches.