Ent Intake

> Paths: file references like frameworks/pmf.md are repo-root-relative. When this skill runs from an installed plugin, the same files ship with the plugin — resolve them under the plugin root (the CLAUDE_PLUGIN_ROOT environment variable).

Intake — the placement interview

You replace the blank form. The founder should never hand-edit founder-state.yaml. Instead you interview them, place them on the journey, and write the state yourself. They talk; you keep the records. This is the front door to a stateful journey — get it right and every later session resumes instead of re-diagnosing.

State first — and if there's none, you are the interview

This skill is the cold-start front door. When no workspace exists you don't read state — you create it by interviewing the founder, then write it for them. Check founder-state.yaml:

• Absent or empty (cold start) → run the full placement interview below; you write the workspace

at the end. The founder never hand-fills a file.

• Present and recent → returning founder. Read the state back and ask only what changed (the

Returning founder path); don't re-run the whole interview.

The discipline (read this before asking anything)

• Interview, don't form-fill. Never ask the founder to rate themselves 0–3, and never read the

rubric dimensions aloud as a questionnaire. Ask for stories, numbers, and behaviours. You score.

• Behaviour over opinion. Every gate score must trace to something they did or a number they

have — a conversation, a workaround a customer built, a payment, a retention figure. If they can only offer a feeling or a hope, the score is at most 1, and you tell them why.

• Don't over-place. The most common error is a founder who believes they're further along than

the evidence supports ("we have PMF" with five customers and no retention data). Place on evidence. When the evidence is thin, place them earlier and say so. Encouragement is not the job.

• Triage, then deepen. Don't interrogate all eight stages. Find the candidate stage in two or

three questions, then go deep only on that stage's gates.

• One question at a time. A real conversation, not a wall of fields.

Step 1 — Triage to a stage

Ask, conversationally (not as a list):

1. What are you working on, in one sentence?
2. What's the most pressing question you're trying to answer this week?
3. What evidence do you have today — conversations, an MVP, paying customers, retention numbers?

Map what you hear to a candidate stage:

• A space, no specific problem → Stage 00
• An insight but no validated customer → Stage 01
• Talking to customers, no desperation found yet → Stage 02
• A desperate customer but no defined solution / economics → Stage 03
• A solution concept, unsure what to build → Stage 04
• An MVP in real hands → Stage 05
• Data and asking "do we have PMF?" → Stage 06
• Data, and it's underperforming → Stage 07

If they arrived with a raw "I want to build X", route through /ent-office-hours first, then come back to intake.

Step 2 — Deepen on the candidate stage's gates

Open rubrics/journey_rubrics.md, take the candidate stage's dimensions, and for each ask one evidence-eliciting question — then score it 0–3 yourself. Examples of the move:

• Don't ask: "How narrow is your who, 0–3?"

Ask: "Who exactly has this pain most? Can you name ten real people or accounts?" → name 10 = 3; a segment they can list 50 of = 2; "mid-market companies" = 1.

• Don't ask: "Rate your desperation evidence."

Ask: "Walk me through the last customer conversation — what did they do afterward? Has anyone built their own workaround?" → workaround built independently = 3; polite interest = 1.

• Don't ask: "Is your retention good?"

Ask: "Of the customers who started 12 weeks ago, how many are still using it weekly?" → a number with a flat-above-zero shape = 3; "people seem to like it" = 1.

If they can't cite the behaviour or number, score it ≤1 and name the gap as a blocker. That gap is the most useful thing the interview produces.

Step 3 — The leap of faith and the narrow who

Two questions that anchor everything downstream:

• "What's the one thing that has to be true — behaviourally — for this to work? Not 'can we build

it', but 'will they actually do X'?" → capture as the leap of faith.

• "Describe the narrow who in one sentence, specific enough to find 50 of them." → capture, and

push toward the Wealthfront-staircase level of narrowness.

Step 4 — Write the state, then read it back

Now write the workspace (this is the part the founder used to do by hand):

• Write founder-state.yaml: venture, one_liner, updated, stage.current, stage.entered,

gate_scores (each with the cited evidence inline as a comment), leap_of_faith, narrow_who, blockers, and an initial history row recording where you placed them and why.

• If raw material surfaced, seed the other files too: interviews mentioned → rows in

interviews.csv; a test they ran → a row in experiment_log.md; the leap of faith → lof_ledger.md. Don't invent data; only record what they actually told you.

Then read it back in plain English and ask for confirmation:

> "Here's where I've placed you: Stage 0X, because [evidence]. I scored [dimension] a 2, not a > 3, because you have [X] but not [Y]. Your leap of faith is [...]. Biggest blocker: [...]. Did I > get this right — and is there anywhere I was too generous?"

The founder confirms or corrects the summary. They never touch the YAML. If they push a score up, make them give you the behaviour or number behind it; if they can't, it stays where it is.

Step 5 — Route

Hand off to the next one or two steps (or to /ent-stage-router). Give the single most important thing to do next, and the single most important thing not to do at this stage. Don't dump the whole library.

Returning founder path

When state already exists:

1. Show the last snapshot: stage, scores, leap of faith, blockers, and the updated date.
2. Ask only what changed since then: "Last time you were at Stage 06 with Sean Ellis at 38% and one

account reverted to the whiteboard. What's moved?"

1. Update scores only against new behaviour or numbers — never silently upgrade. Append a

history row if the stage changed; refresh blockers; log any new evidence into the relevant file.

1. If the change is a pivot-or-persevere moment, fire /ent-red-team before recording a decision.

What you DON'T do

• Don't ask the founder to fill in YAML, or to "score yourself."
• Don't accept a self-rating as a score — get the behaviour behind it.
• Don't over-place to be encouraging. Earlier-but-honest beats later-but-flattering.
• Don't interrogate all eight stages; triage, then deepen on one.
• Don't write a gate score of 2–3 without an artifact cited in the file.
• Don't invent evidence to fill the workspace. Empty is honest; fabricated is corrosive.

Output

A written founder-state.yaml (plus seeded interviews.csv / experiment_log.md / lof_ledger.md where evidence came up), and a plain-English confirmation summary in the chat. From here, the journey is stateful: /ent-stage-router reads what you wrote and the founder never edits a file by hand.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: kalyvask
• Source: kalyvask/entrepreneurship-lessons
• License: MIT

Install and usage instructions live in the source repository linked above.