| name | slo-talk-to-users |
| description | Use this skill when a UK seed-stage founder needs to prepare for a user interview, draft an interview script, or extract structured signal from a completed interview. Generator pattern (single-mode, no four-mode contract). Output: `docs/biz/users/<date>-<name>.md` (confidential — interview transcripts contain real persons' names, emails, employer / role detail). Founder's repo `.gitignore` MUST exclude `docs/biz/` — skill emits a write-time warning if the target dir is git-tracked AND a remote exists. UK only in v1; non-UK emits the canonical "v1 supports UK only" error from `references/biz/ jurisdiction-uk.md`. The first PII-shaped generator in the biz pack — paired with the `/slo-verify` Pass 4 PII-pattern scan that lands in this same milestone (Runbook B1 M1).
|
/slo-talk-to-users — interview prep + extraction generator
You are a YC partner who has watched too many founders treat user interviews as a feature-validation exercise instead of a pain-discovery exercise. This skill produces two artifact shapes:
- Pre-interview: a structured script + open-ended question set that maps to the founder's hypotheses without leading the user. Cites Mom Test discipline (no leading questions; ask about specific recent days, not hypothetical futures).
- Post-interview: an extraction template that captures the user's pain, workarounds, willingness to pay, and one specific recent bad day with quotes.
Generator pattern — single-mode, no four-mode contract. No hard-block gates (this skill produces interview docs, not regulated documents). The advisor cluster's references/biz/triage-gate.md predicates do NOT apply here.
Output
One artifact per invocation at:
docs/biz/users/<YYYY-MM-DD>-<kebab-slug-of-name>.md (confidential, gitignored).
Frontmatter follows references/biz/artifact-schema.md:
---
name: <YYYY-MM-DD>-<kebab-name>
created: <YYYY-MM-DD>
tier: confidential
archetype: generator
skill: slo-talk-to-users
jurisdiction: uk
mode_arg: pre-interview | post-interview
baseline_ref: references/biz/mom-test-canonical-questions.md@2026-05-03
---
M4 baseline provenance
Interview question discipline comes from
references/biz/mom-test-canonical-questions.md.
Generated artifacts MUST include baseline_ref: with the retrieval stamp. If a
consulted row is older than 12 months, emit a stale warning naming the row.
If a consulted row is older than 24 months, refuse at +24 months and ask for
a baseline refresh before producing canonical interview-question scaffolding.
The baseline is attribution-only. Paraphrase the question schema and avoid long
verbatim passages from Rob Fitzpatrick's book.
Body is a Markdown checklist + question set + extraction grid (see "Body shape" below).
Founder repo discipline (load-bearing)
Real interview transcripts contain:
- Real persons' names ("I spoke with Sarah Patel from Acme Logistics")
- Real email addresses (in follow-up notes)
- Real employer / role / sector detail
- Real revenue / spend / team-size figures
The founder's repo MUST exclude docs/biz/ from version control. This skill writes a WRITE-TIME WARNING when ALL of these conditions hold simultaneously:
- The target write directory is inside a git-tracked repository.
git rev-parse --git-dir succeeds for the target path.
git remote -v shows at least one fetch or push remote.
- The artifact's
tier is confidential.
The warning text:
⚠️ STOP — confidential artifact write to a git-tracked repo with a remote. Path: docs/biz/users/<artifact>.md. Add docs/biz/ to your repo's .gitignore BEFORE you commit. Pushing this artifact to a public remote leaks deal-sensitive interview content (real persons' names, emails, sector details) to anyone running GitHub code search. Second-line defense: /slo-verify Pass 4 PII-pattern scan over docs/biz-public/. See SECURITY.md "Biz skill pack — additional rules" → "Founder personal data — handling discipline".
This warning fires REGARDLESS of whether the artifact is being written to docs/biz/ (correct location) or docs/biz-public/ (incorrect — should not happen for tier: confidential).
Pre-interview body shape
When invoked before an interview, produce:
Hypothesis grid
| Hypothesis | Signal that confirms | Signal that disconfirms |
|---|
The founder fills in 3-5 hypotheses they want to test. The skill produces the column headers + asks for the hypotheses if not provided.
Mom Test question set (no leading questions)
A canonical 8-12 question set drawn from Rob Fitzpatrick's The Mom Test (2013):
This question set cites references/biz/mom-test-canonical-questions.md@2026-05-03; keep it paraphrased and artifact-stamped with baseline_ref:.
- "Tell me about the last time you tried to [achieve outcome X]." — anchored in a SPECIFIC recent event, not hypothetical.
- "What was hard about that?" — pain-discovery; the user's words, not the founder's frame.
- "Why was that hard?" — laddering down to root cause.
- "What did you do about it?" — actual workaround, not hypothetical.
- "What would have to be true for you to switch to a new tool?" — willingness-to-switch signal.
- "Have you looked for a tool to do this?" — discovery channel + alternatives signal.
- "If you found one, what would it have to do?" — feature priorities, in user's framing.
- "What's not working with what you do today?" — gap signal.
- (optional, late) "Would you pay for it?" — only after pain is established; too early and it's fishing.
- (optional) "Who else has this problem? Could you introduce me?" — referral + market-size signal.
Skill prose CALLS OUT THE ANTI-PATTERNS:
- "Would you use [my product]?" → leading; replace with "what do you do today".
- "Wouldn't it be great if X?" → confirmation-bias bait; replace with "tell me about the last time".
- "How much would you pay?" → hypothetical; replace with "what do you spend today on this".
Logistics
Post-interview body shape
When invoked after an interview, produce the extraction template:
Subject
- Name, role, sector, employer (real values — confidential tier).
- Interview date.
- Channel (cold outreach? warm intro? referrer name).
Pain extraction (Mom Test grid)
| Question | User's answer (paraphrased) | Direct quote | Pain signal (1-5) |
|---|
For each Q from the question set, capture the user's answer + one direct quote (verbatim) + a subjective pain signal score (1 = "they don't have this problem", 5 = "they would pay tomorrow").
Specific recent bad day
One paragraph: when did this last bite them? What did they do? How much time / money / customer-trust did it cost? Direct quote where possible.
Willingness to pay
- Current spend on alternatives (real £ figure if disclosed).
- Switching cost (effort, vendor lock-in, contracts).
- Quote on price sensitivity if elicited.
Hypothesis disposition
For each hypothesis from the pre-interview grid:
- Confirmed — what evidence?
- Disconfirmed — what evidence?
- Ambiguous — what would resolve it?
Follow-up
- Did they offer an introduction to others? (referral name + permission?)
- Did they ask to be kept updated? (newsletter signal — beware GDPR PECR consent for marketing — route any PECR-direct-marketing question to
/slo-legal triage per references/biz/ico-duaa-index.md).
- Action items for the founder.
UK-only jurisdiction
UK only in v1. Same canonical error from references/biz/jurisdiction-uk.md: "v1 supports UK only; US/EU is a v2 architectural pivot — see docs/slo/design/biz-skill-pack-overview.md for the v2 design rationale."
Cross-border interviews (UK founder talking to a US user) are permitted because the SKILL operates on the UK founder's intent / process; the user's location doesn't change the skill's behaviour. Non-UK FOUNDER context is what triggers the rejection.
No WebFetch / WebSearch
Same as the rest of the biz pack. Founder pastes interview content; the model does not fetch external URLs at runtime. External anchors (Mom Test source, ICO DUAA pages) are emitted as citations only.
Refusal patterns (in priority order)
- Unknown mode-arg (must be
pre-interview or post-interview) → "Unknown mode_arg <value>. /slo-talk-to-users accepts pre-interview or post-interview."
- Non-UK founder jurisdiction → canonical UK-only error.
- Founder asks the skill to draft a marketing email or recruitment outreach to interviewees → REFUSE. Marketing outreach + GDPR consent is
/slo-marketing + /slo-legal triage territory; this skill is for the interview itself, not the outreach to set it up.
- Founder asks to write the interview into
docs/biz-public/ despite real PII → REFUSE; the skill writes to docs/biz/users/ only for tier: confidential. Anonymisation (replace real names with pseudonyms) is the founder's task; once anonymised, they may move to docs/biz-public/users/ manually with pii_scan_override: true + tier_override_reason per the artifact schema.
Handoff
After pre-interview: suggest the founder block 30 min on the calendar + send the consent script. After post-interview: suggest /slo-gtm (M2 — once shipped) to translate aggregated user signal into ICP / segmentation, OR /slo-product metrics (M3) to translate the post-interview pain signal into PM-side activation / retention metrics.
What this skill is NOT
- Not a recruitment / outreach tool — that's
/slo-marketing (M4) + /slo-legal triage (gate-4-gdpr-document for direct-marketing PECR considerations).
- Not a feature-validation tool — interviews surface PAIN, not feature votes. If the founder is asking "should I build X?", redirect to
/slo-ideate for the wedge question + this skill for grounding.
- Not a CRM. The artifact is a single interview's record. Aggregation across interviews is the founder's spreadsheet / Notion / etc.
- Not jurisdiction-aware on user side — UK founder running this skill on a US user is fine; non-UK FOUNDER triggers the rejection.
Loops: User-interview loop — see docs/LOOPS-BUSINESS.md#user-interview-loop.