agentleFS
AhmedQureshi89AhmedQureshi89/lean-experiment-kit/.claude/skills/get-started/SKILL.md

get-started

Skill0 starsChanged 19 days agoMITView on GitHub

The kit's first run. Use when the founder types /get-started or says "set me up", "how do I start", "onboard me", or opens a kit that still holds the example business - a short welcome, an optional tour, a few questions about their business, then their own config, segments and two canvases written, the example removed, a clean build, the dashboard created, and one first test proposed.

26 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 1 is distinctive, and 5 are very specific to this project.

  • 0

    Common

    Most of the 25 edited versions we read keep these.

  • 1

    Distinctive

    This author's own choices. Worth a look.

  • 5

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read kit.config.json, customer-segments.md and the two headers under experiments/canvas/.

    line 18
  2. Distinctive:

    Ask in this order, one group per message.

    line 46
  3. Very specific:

    After each answer, ask "how do you know?" once for the group, in plain words: *I've seen it in real data or real customers told me* → seen.

    line 56
  4. Very specific:

    Leave experiments/PROGRESS.md alone: the build writes it.

    line 111
  5. Very specific:

    Run python dashboard/build_data.py (python3 on a Mac).

    line 122
  6. Very specific:

    Run /sync-dashboard. On the first run it creates the founder's private dashboard and writes its link into kit.config.json.

    line 128

“Common” is measured: kept by at least 6 in 10 of the 25 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Take a founder from a fresh copy of the kit to their first test card in about fifteen minutes. They answer questions in plain words; Claude writes every file. Nothing is sent, posted or pushed.

How to ask. Two to four questions per message, never a form. Give a one-line example answer for each, taken from the Bean Box example, so the founder sees the size of answer wanted. "I don't know" is always a good answer: it becomes an honest assumption with Evidence 0. Use AskUserQuestion where the answer is a choice (the weekend, tour or set up now); plain chat otherwise. Never ask for a customer's name, email or phone number: contact lists come later, in a CSV.

0. Where the kit is

Read kit.config.json, customer-segments.md and the two headers under experiments/canvas/.

  • Not set up yet: business ends with (example), or a canvas title ends with (example). Go on.
  • Set up already: neither holds. Say so in one line, offer /tour of their own files and the four phrases of step 9, and stop. /get-started never overwrites a founder's own canvases. To redo the setup, they say "start over", and only then go on.
  • Half done (for example, the config is theirs but a canvas still says (example)): say which parts are done and go on from the first part that is not.

1. Welcome (one short message)

Say what the kit does in three lines, show the loop, say what they will have at the end, and ask one question:

  Hypothesize → Experiment → Learn → Decide (persevere / pivot / kill)
       ↑                                            │
       └──────────── one next hypothesis ───────────┘

At the end they will have their business on two canvases, the riskiest belief picked out, a private dashboard, and a first test ready for their go. Ask: see how it works on the example first (/tour, about five minutes), or set up my business now? If they pick the tour, run /tour and come back to step 2 when it ends.

2. The questions

Ask in this order, one group per message. Write down each answer with how they know it.

GroupAskBecomes
A. The businessIts name. What it does, for whom, in one sentence.business; the canvas titles and segment lines
B. Where you workYour city or time zone. Which days are your weekend. Which day your week starts. A GitHub link for this repo, if you have pushed it (optional).timezone (an IANA name such as Europe/Berlin), utc_offset (today's offset, such as +01:00), weekend (["Sat", "Sun"]), week_starts, repo_url
C. Your first customersWho exactly you are building for first (Seg 1). Anyone after them (Seg 2, optional). For Seg 1: the job they are trying to get done, the worst part of it today, and what they wish happened instead.customer-segments.md; VP-A J, P, G; LC-A PR, CS
D. Your offerWhat you sell or will sell. How it takes away that worst part. How it gives them what they wish for. What is already built and working, and what is not.VP-A S, R, C; LC-A UV, SO
E. The modelWhere you will find these customers. How you will make money, and roughly what price. Your main costs. The one number that would tell you it is working. Anything that is hard for others to copy ("nothing yet" is fine).LC-A CH, RS, CO, KM, UA

After each answer, ask "how do you know?" once for the group, in plain words: I've seen it in real data or real customers told me → seen; one person said it, or I'm inferring it → signal; it's my belief → assumption. For the business side of VP-A (S, R, C) and LC-A (SO, CH, CO, KM, UA) the tags are live (built and running), untested (exists or is claimed, no proof it helps) and gap (missing).

3. Score it together

Show the rows as one table in chat, with your proposed Matters (1–3: how much the business depends on it; 3 = if this is wrong, there is no business) and Evidence (0–3: 0 belief, 1 one voice or a hint, 2 seen once in real data or tried once, 3 established). Explain the two scales in one line each, then ask them to correct anything.

Be honest on their behalf: Evidence 2 or more only for something seen in real data or done once, never for a strong feeling. Most rows of a new business are assumption, Evidence 0, and say so kindly: that is what the experiments are for.

4. Confirm before writing

List what will change and ask for a go:

  • kit.config.json: name, time zone, weekend, week start, repo link. A dashboard_url already there (from a tour that created the dashboard) is kept: the same dashboard is reused.
  • customer-segments.md, experiments/canvas/VP-A.md, experiments/canvas/LC-A.md: rewritten with their answers, version 1.
  • The example experiments experiments/0001-work-from-home-interviews/ and experiments/0002-subscription-mock-sale/ deleted. Only delete a folder whose test card is the Bean Box example; anything else is the founder's and stays. Git keeps the example if they want it back.

5. Write the files

Rewrite each file in place: keep every header key, the "How this file works" block, every ## heading and every table's header row exactly as they are, because the build and the dashboard read them. Replace the rows. Remove the example banner (the first > block that says it is a made-up example) and the "Example only" note.

  • kit.config.json: the answers from group B. If they gave a city, choose its IANA zone and today's offset, and say which you chose.
  • customer-segments.md: one row per segment (Seg 1, Seg 2 …) with Evidence so far ("None yet" when it is belief). Keep the "Known limits" section.
  • experiments/canvas/VP-A.md: header version: 1, date today, title: <business> Value Proposition Canvas, segment (Seg 1 in words), summary (two sentences: what is seen, what is belief), riskiest and riskiest_why (step 6). At least one row of each kind: J, P, G (customer side), S, R, C (business side), numbered J1, J2 … The Fit, read by pair table gets one plain line for each of its three pairs. The Facts table keeps its header row and holds a row only for a real number they gave. Changes: | 1 | <today> | First draft from the founder's answers in /get-started. |
  • experiments/canvas/LC-A.md: the same header keys (keep kind: lean-canvas and zoom_in: VP-A), title: <business> Lean Canvas. At least one row in every one of the nine blocks: PR, CS, UV, SO, CH, RS, CO, KM, UA. PR1 is the problem in one line and SO1 the solution in one line: the detail lives in VP-A. Keep the Read by theme table's three rows naming the blocks (Desirability (PR, CS, UV), Feasibility (SO, CH, UA), Viability (RS, CO, KM)), with one plain line each, so every row sits in one theme. Same Changes line.
  • Delete the two example folders.
  • Leave experiments/PROGRESS.md alone: the build writes it.

6. Pick the riskiest box

In each canvas, the riskiest row is the one with Matters 3 and the lowest Evidence. On a tie, take the customer side first (Desirability before Feasibility before Viability), then the row everything else depends on. Write it to riskiest, with riskiest_why in one or two plain sentences. Tell the founder which rows you picked and why, and let them overrule you.

7. Build until clean

Run python dashboard/build_data.py (python3 on a Mac). Each PROBLEM: line names the file and the row: fix it and run again until the build prints the canvases with no PROBLEM: line. Do not show the founder the problems you fixed unless one needs their answer.

8. The dashboard

Run /sync-dashboard. On the first run it creates the founder's private dashboard and writes its link into kit.config.json; later it updates the same one. Tell them to approve the uploads when asked, give them the link, and suggest they open it on their phone and keep it there. It is private to them until they share it.

Then commit the setup, naming the files (never git add -A): kit.config.json, customer-segments.md, experiments/. Message: Set up the kit for <business>. Push only if they ask.

9. The first test, and the four phrases

Propose exactly one first hypothesis against the riskiest box, in the form "We believe that <Seg 1> <the belief>", with the cheapest method that gives real evidence (usually #1 Customer Interview for a pain or a job nobody has seen yet). Ask: "Shall I write the test card?" On a yes, run /new-experiment.

End with the four things they will say from now on:

SayWhen
"test the idea that …"you have a belief to test
"pull the numbers for 0001"a test is running and you want today's count
"log 0001"its window has closed and you are ready to decide
"sync the dashboard"rarely: Claude does it at the end of each change

And: /tour walks through their own files any time.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.