PM - Discovery Interview Prep

Agent mode (--agent)

Supports --agent: runs autonomously in a subagent, drafts the agenda from existing inputs, and returns a short summary + coverage note.

• No flag → interactive (default); if inputs are heavy, offer agent mode.
• --agent → obey. First check inputs are complete. Anything missing: do NOT invent it - mark [ASSUMED - what/why] in the output and summary. Never hallucinate to fill a gap.

Standalone run

Runs with or without a Pureinn workspace - it needs no other skill's artifact, so it works as a single tool from an install that never ran /pureinn.

• No workspace: create just the folder this skill writes into, or write to a path the user names, and say where the file went. Do not scaffold a project, do not invent a state.json, and do not send the user to /pureinn first.
• Read pureinn-variables.md / state.json where a value is actually used, not at the top of the run. Missing value: continue and name the capability it costs.

What this skill does

Prepares you for a live discovery session. It does NOT run the interview - the interview happens in the room, guided by the agenda this skill produces. This skill:

1. Reads what discovery has already covered (prior meetings/ notes, Track A-D artifacts, the Discovery Report if one exists)
2. Builds a coverage map - which discovery areas are done, thin, or untouched
3. Compiles a session agenda targeting the biggest gaps, with the actual questions, follow-ups, and techniques written out in full
4. Saves the agenda to meetings/prep/ so it pairs with the meeting note that pm-meeting produces afterward

Self-contained (no external file needed): the complete question bank and technique library are embedded in this skill, in the Question bank section at the bottom. The skill depends on nothing outside itself - no companion document, no docs/ file. Run it standalone at any time: to compile a targeted agenda, to walk the full set of questions live with a client, or to proof-check your own idea by being your own toughest client.

Standalone agenda output: the agenda it produces is also self-contained - every question, follow-up, red-flag answer, and technique is written out in full, not referenced by module number. Print the agenda and walk into the session with nothing else; no laptop, no file, no Claude running during the session itself.

The coverage areas (module map)

Each area maps to a module in the Question bank below. Use this to decide what a session should target; pull the actual questions from the matching module.

Step 0: Coverage check (current state)

Read, in this order (deep source ingestion - full files, not skims):

1. meetings/ - all prior client-discovery notes (and customer-discovery if users were interviewed)
2. meetings/prep/ - prior agendas (what was planned but maybe not covered)
3. artifacts/phase-2-discovery/ - Track artifacts + discovery-report.md if present
4. state.json / assessment - project context

Build the coverage map: per area above, mark ✅ covered / ⚠️ thin / ❌ untouched, citing where the evidence lives. Confirm coverage: "Read [N] meeting notes + [N] artifacts."

First session (nothing exists yet): normal case, not a blocker. Default agenda = M0 research summary + M1.1-1.3 + M1.8 (trigger, business, success, their user claims) - relationship-building first, constraints sweep second session. Note it as such.

Interaction: Group related questions (2-4 per round) and confirm before moving on. For any A/B/C/D choice, use the AskUserQuestion tool with one option marked (Recommended) - never print options as plain text. Keep open-ended questions free-text (don't fake options). If the user is unsure, propose 3-4 concrete options plus "Other". Surface an assumption the moment you make one; never fabricate to fill a gap. (Full standard: CLAUDE.md.)

Step 1: Session parameters

Use AskUserQuestion tool for these two questions:

Who is this session with?

A) The commissioner - client / sponsor / exec (Recommended if Plane 1 gaps remain - their context anchors everything else) B) Their staff - the people who will operate the product daily C) Their end users / customers D) Mixed - commissioner plus users or staff in one room

What format and length?

A) Video call 30-45 min - focused, 1-2 areas max B) Video call 60-90 min (Recommended for discovery depth - fits 2-3 areas plus close) C) On-site workshop half-day - process walkthroughs possible (M2b works best live) D) Async / written - questionnaire, use only for constraint sweeps, never for trigger/JTBD

Then ask as plain text:

Date of the session, and is there anything specific you already know you need from it? (e.g. "they keep mentioning a competitor's app", "I need budget clarity before quoting")

If the user is unsure what the session should target, recommend from the coverage map: the highest-leverage gap given where the engagement stands (early → trigger/business; before quoting → budget/constraints/scope; before spec → processes/exceptions).

After answers, confirm the session focus, then generate.

Step 2: Generate the agenda

Generate in English. Select 2-3 areas maximum (per hour of session) from the coverage gaps - depth beats coverage. Pull the matching questions from the Question bank section below and write them out in full in the agenda - question text, follow-up, red-flag answer, and (where relevant) the complete technique steps, not a "see module X" pointer. Adapt every question to this client's context (their industry, their words from prior notes, their named references). Generic questions are a waste of the session - personalize every one.

Self-containment rule: the agenda must be usable by someone who has never seen this skill and will not have anything open during the session. If a technique applies (e.g. Reference decompression), copy its steps into the agenda block in full, condensed to what fits the moment - not a name-drop.

ARTIFACT: Session Agenda

# Discovery Session Agenda - [date] - [audience]

> **Session [N]** · **With:** [names/roles] · **Format:** [call/workshop, length]
> Self-contained - everything needed for this session is below. No other document required.

---

## Coverage status going in

| Area | Status | Source |
|---|---|---|
| [area] | ✅ / ⚠️ / ❌ | [meeting note / artifact] |

**This session targets:** [the 2-3 gap areas chosen and why they're the highest-leverage now]

---

## Opening (1 min)

[Suggested framing in one or two sentences - why we're here, what we'll walk out with. Personalized, not generic.]

## Block 1 - [Area name] ([X] min)

Goal: [what must be known by the end of this block]

- [Question 1 - adapted to this client's context]
  - follow-up: [the peel-the-onion follow-up]
  - red-flag answer: [what answer signals a problem, and what to do next]
- [Question 2...]

[If a technique applies this block, write its steps out here in full - e.g.:]
**Technique - Reference decompression** (they mentioned [XY] in session 1, likely to come up again):
1. Split the layers: "What exactly about [XY] - how it looks, how it flows, or what it lets people do?"
2. Force a scenario: "Walk me through how you'd actually use that here."
3. Find the job: "What would that solve for you that you don't have today?"
4. Check fidelity: "Have you used it yourself, or seen it somewhere?"
5. Iceberg move: "Under the hood, [XY] also does A/B/C to make that work - which of those do you need?"
6. Outcome test: "If we got you the same result a different way, would that work?"
7. Classify before moving on: **directive** or **hypothesis** - write it down on the spot.

## Block 2 - [Area name] ([X] min)

[same structure]

## Close ([~10] min)

- Recap: "Here's what I heard: ..." - confirm or correct
- Next steps: ours + THEIRS (content, access, user intros - name what they owe)
- Next session target: [what the following session should cover]

---

## Bring / prepare

- [prior notes to re-read, materials from client to review - this agenda is everything else you need]

## After the session

Run `/pm-meeting` on the notes same day (Client Discovery type) → then `/pm-discovery-report` to update the running report.

Internal completeness checklist

• Coverage map built from actual files, not assumed
• 2-3 areas max per hour - agenda is not a checklist dump of the whole question bank
• Every question personalized to this client (their industry, their words, their references)
• Each block has a goal, follow-ups, and at least one red-flag answer
• Techniques matched to expected session content (references expected → decompression ready)
• Close block present with recap + mutual next steps
• If mixed audience: commissioner questions and user questions separated (never probe budget in front of staff)

Save to

pureinn-workspace/[project-slug]/meetings/prep/[YYYY-MM-DD]-[audience-slug]-agenda.md

Question bank (complete - the skill's source, embedded)

Everything needed to compile an agenda or walk a session live. Two planes (the commissioner above, the real users below), three user populations (the client's customers, the client's staff, the client's own oversight). Provenance rule throughout: real user research > [CLIENT-ASSERTED] (client's claim about their users) > [ASSUMED]. Never let a client's claim silently become a validated persona.

Module 0 - Pre-meeting research (before the first session)

• The client's company: what they sell, to whom, how they position themselves
• Their market and their competitors (in commissioned work, the market lens is the client's market, not yours) → /pm-market-analysis, /pm-domain-analysis
• Their current digital footprint: existing site/app/system, visible workflows, reviews of their service
• The people attending: roles, likely priorities, likely objections
• Any prior materials they sent (brief, RFP, deck) - read fully first
• Output: 3-5 hypotheses about their likely problems and priorities, to test in Module 1

Module 1 - The commissioner (Plane 1)

1.1 Trigger and why-now → Discovery Report (context), /pm-project-charter

• What brought you to this? What made you take this meeting? (the real trigger - get them talking early)
• Why now - why not a year ago, why not next year? (urgency, external pressure)
• What happens if you do nothing for the next 6-12 months? (is the problem fund-worthy; consequence framing)
• Have you tried to solve this before? What happened? (prior attempts, why past solutions failed)

1.2 Their business and the product's job in it → Discovery Report, /jtbd-building (commissioner's job)

• Walk me through your business - how do you make money, who pays you? (the business model the product serves)
• Where does this product fit? What does it have to do for the business? (job at business level: earn more / spend less / reduce risk / comply / status)
• Which part of your operation does this touch or replace? (integration surface → Module 2b)

1.3 Benefit, success, alignment check → Discovery Report (value hypothesis), Scope Brief (acceptance)

• How will you measure whether this was worth it? What number moves? (success in their terms)
• What KPIs are you responsible for? (personal stake)
• When would you call it a failure? (the inverse test - a criterion that cannot fail is not one)
• Alignment check (explicit): compare the client's benefit with the users' benefit (Module 2). Client wants control/reporting/upsell, users want simplicity/speed → surface the tension as a finding, don't design around it silently.

1.4 Budget and decision process (never open with "what's your budget?") → Meeting notes, Discovery Report

• How are you determining the value of solving this? What's it costing you today? (value-first: the number surfaces through impact)
• Have you solved something like this before - what did that cost? (anchors from their history)
• When you think about solving this, is there a range in mind, or are we still shaping that? (direct but calm; an out that still gives signal)
• Besides cost, how will you choose between options? (decision criteria; where you differentiate)
• Who else needs to sign off, and what does that process look like? (approval chain, procurement, timeline reality)

1.5 Stakeholders and the late-veto check → Meeting notes; /pm-stakeholder-map if corporate-sized

• Who on your side works with us day to day? Who decides? (contact vs. decision-maker - usually two people)
• Who can block or kill this at the END, close to launch? (IT, legal, co-owner, key franchisee) (the late vetoer - get them an early touchpoint)
• Whose work does this change? Who might resist it? (internal adoption risk)

1.6 Constraints sweep (each one skipped is a surprise later) → Scope Brief, /pm-tech-feasibility, /pm-domain-analysis

• Integrations: what systems today - ERP, CRM, accounting, payments, inventory, booking? Which must connect? Who owns them?
• Data: what exists, where does it live, quality, migration needed, who owns it legally?
• Content: who supplies texts, photos, product data, translations, by when? (the classic web-project killer - never assume they'll "just deliver content")
• Brand / design: existing brand book, design system, non-negotiable visual identity?
• IT / security: hosting, IT policy, security review, data residency?
• Operations after launch: who maintains and administers this, what's their technical skill, support model?
• Replacement / migration: does this replace something? What happens to current data, users, SEO, integrations? Parallel run or hard switch?
• Time / budget: hard deadline (event, contract, season)? What flexes first - scope, time, or budget?

1.7 References - "I want it like [company XY]" → run Reference decompression (Techniques below); classify directive/hypothesis before moving on

1.8 Claims about their target group → /pm-personas tagged [CLIENT-ASSERTED]

• Who will actually use this? Describe a real, specific person. (their model of their users)
• How do you know? Have you talked to them about this? (evidence vs. assumption)
• Who is your best customer today - and your most painful one? (segment texture)

1.9 Client maturity (decides whether Plane 2 is an import or your job)

• Have you done research with your users - interviews, surveys, analytics? → yes: import (Path A of Track skills) / no: run Plane 2 yourself or proceed [CLIENT-ASSERTED] with eyes open
• Do you have analytics on the current solution? → yes: baseline for Scope Brief / no: note the blind spot
• Previous attempts, agencies, prototypes? → read the postmortem, cheapest lesson available

Module 2 - The real users (Plane 2)

2.1 Three populations - always sweep all three (staff personas are legitimate personas → /pm-personas)

• a) Customers: who are they, what job are they hiring this for? (nobody skips this one)
• b) Staff: who operates this day to day on your side? What's their workday? (skip it → staff can't operate the launch)
• c) Client's own view: what do YOU need to see - reports, oversight, control? (skip it → the person who paid never sees value)

2.2 Segments, jobs, pains → /pm-personas, /jtbd-building, /pm-problem-validation

• Which type of user/customer matters most? Which hurts most today? (segment priority by pain intensity)
• Walk me through the last time [user] had to deal with [the problem]. (concrete episode - Mom Test; stories over opinions)
• How often does that happen? In what situation - where, when, under what pressure? (job frequency and context)
• What do they do about it today? Tools, workarounds, spreadsheets? (the real competition)
• What's the most frustrating part? What have they given up on? (pain intensity, resignation)
• What would change for them if this worked perfectly? (expected gains, outcome language)

2.3 Empathy sweep (depth on a key persona) → /pm-personas Six quadrants - fill each: think and feel (worries, aspirations), say and do (public behavior, contradictions), see (environment, competitors' offers), hear (influencers, colleagues, reviews), plus pains and gains.

Module 2b - Process walkthrough (for any "system / workflow / portal" request)

Seeds domain modeling → /pm-process-flows, /pm-entity-registry; pains → /pm-problem-validation. Pick one end-to-end process (an order, a booking, a claim, a publication):

1. Walk it start to finish: "take me through what happens from the moment X arrives until it's done." Let them narrate; don't interrupt with solutions.
2. Capture per step: who does it (actor), what they touch (system, paper, phone), what they hand off and to whom, how long, how often (volume).
3. Mark the friction: where does it break, wait, get retyped, get lost?
4. Run the Exception sweep (Techniques below) before leaving the process. Repeat for each core process (usually 2-4 in a commissioned build).

Module 3 - Solution and scope shaping (only after M1-M2b) → /pm-scope-brief

• Of everything we discussed - what MUST v1 do, or it's not worth launching? (the real core, their words)
• What would be great later but doesn't block launch? (deferred scope - captured, not lost)
• What should this explicitly NOT do or be? (non-goals - cheapest scope protection)
• If we achieved the same outcome a different way than you imagined, is that acceptable? (outcome vs. output - design freedom or literal directive)
• How will you evaluate that we delivered what you wanted? Who signs off? Demo cadence? (acceptance model)
• Re-run the Exception sweep against the proposed scope.

Module 4 - Close

• "Looking at the clock, ~10 minutes left - let's align on next steps." (interject professionally; never let it just run out)
• Recap: "Here's what I heard: [3-5 points]. Did I get that right? What did I miss?"
• Concrete next steps: who, what, by when - including what THEY deliver (content, access, user intros).
• Same day: run /pm-meeting (Client Discovery type) on the notes.

Techniques

Reference decompression ("like company XY") - run every time a reference appears:

1. Split the layers: "What exactly about [XY] - how it looks, how it flows, or what it lets people do?" (visual / UX / function / business model)
2. Force a scenario: "Walk me through how you'd actually use that here." (shallow references collapse)
3. Find the job: "What would that solve for you that you don't have today?"
4. Check fidelity: "Have you used it yourself, or seen it somewhere?" (used / seen / heard-of)
5. Iceberg move: "Under the hood, [XY] also runs A, B, C, D to make that work. Which of those do you actually need?" (best scope protection + strongest credibility signal at once)
6. Outcome test: "If we got you the same result a different way, would that work?"
7. Classify before moving on: directive (→ Scope Brief as a bounded requirement + explicit non-goals) or hypothesis (→ assumptions//pm-hypotheses, never into spec as settled).

Exception sweep - after any process walkthrough or scope discussion. Tag product-enforced exceptions [CANDIDATE-BR] → /pm-business-rules-library:

• Failure edges: cancellation, return, refund, no-show, error?
• Role edges: special permissions? Who can override?
• Volume edges: peak season, bulk operations, everything arriving at once?
• Lifecycle edges: first user, offboarding, account/data deletion (GDPR), dormant accounts?
• Migration edges: existing customers/data/contracts on day one?

Conversation craft:

• Open-ended, then silent. Ask, then shut up - most people break 5 seconds of silence with the detail they withheld.
• Peel the onion: "You mentioned [X] - tell me more." The first telling is incomplete nine times in ten.
• Consequence chain: "What happens then? Who else does it impact? What does it cost the business?" - impact turns a complaint into a case.
• Say it with questions: not "that will fail because ABC" but "a client of ours hit ABC - how does your plan handle that?"
• No feature tours: later, show ONLY what maps to challenges they named.
• Disqualify honestly: if the problem is too small, the budget unreal, or value below cost - say so and stop. Ending an engagement early is a successful discovery.

Handoff

Čo si teraz má: Cielenú agendu na najbližšie discovery sedenie - mierenú na najväčšie medzery, s otázkami prispôsobenými tomuto klientovi, nie generický dotazník.

Ďalší krok: Vytlač alebo otvor agendu a odveď sedenie priamo podľa nej - je samostatná, nepotrebuješ pri nej žiadny ďalší dokument ani Claude. V ten istý deň /pm-meeting (Client Discovery typ) na spracovanie poznámok → /pm-discovery-report na update priebežného reportu.

Môžeš preskočiť ak: Discovery už konvergovalo (coverage mapa je zelená) - choď na /pm-scope-brief namiesto ďalšieho sedenia.