agentleFS
iangogenticiangogentic/venture-lab/.claude/skills/interview-plan/SKILL.md

interview-plan

Skill0 starsChanged 4 months agoMITView on GitHub

Plan the interviews that would validate each decision.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 7 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈7

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Ask what they did: "Walk me through the last time that happened.

    line 39
  2. Distinctive:

    Never ask what they would do, whether they would use something, or what they think of an idea.

    line 42
  3. Distinctive:

    Do not name the solution, or describe it and ask for a reaction.

    line 52
  4. Distinctive:

    Do not presume the pain. "How frustrating is it when the export breaks?" assumes it breaks and that it matters.

    line 53
  5. Distinctive:

    Do not stack two questions into one, and do not offer the options you find plausible.

    line 55
  6. Distinctive:

    Test each question against the participant who thinks this whole area is a non-issue.

    line 56
  7. Distinctive:

    Write both before any interview happens, and write them so a real outcome could fail.

    line 73

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “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

Interview plan

Turn the riskiest assumptions behind these decisions into questions a real person could answer from their own experience — and that could send the decision the other way.

The question

${question}

The decisions

${decisions}

Start from what could be wrong

Everything upstream was desk research. This plan exists to put the decision in front of reality, so build it backwards from what the decision is standing on: its stated reasons, its biggest unknown, its next validation step, and the counter-evidence it decided over.

For each assumption, before writing a single question, answer this privately: what answer would prove it wrong? If no answer a participant could plausibly give would falsify it, the assumption is not testable by interview — sharpen it or leave it out. Every question then traces back to one such assumption. A question that tests nothing is small talk, however interesting.

Ask about behaviour, not opinion

People are unreliable about their future selves and reliable about last week.

  • Ask what they did: "Walk me through the last time that happened. What did you do next?" Then follow the specifics — what it cost, how long it took, who else was involved, what they tried first, why they stopped.
  • Never ask what they would do, whether they would use something, or what they think of an idea. Predictions about hypothetical products are the most common way an interview round produces a confident wrong answer.
  • Money is behaviour too: what they already pay for, what they built instead of buying, whose budget it came from, what they cancelled and why.

Do not lead

A leading question is one whose wording supplies the answer you were hoping for.

  • Do not name the solution, or describe it and ask for a reaction.
  • Do not presume the pain. "How frustrating is it when the export breaks?" assumes it breaks and that it matters. Ask whether it comes up at all, and let them raise it.
  • Do not stack two questions into one, and do not offer the options you find plausible.
  • Test each question against the participant who thinks this whole area is a non-issue. Can they say so comfortably? If your wording makes that awkward, it leads.

Who to talk to

Describe participants so a recruiter could screen for them: role, context, and a recent behaviour they must have exhibited. A behavioural screener ("has run this reconciliation at least twice in the last quarter") separates people who know from people who will improvise. Say where to find them concretely — the community, the job title to search, the event, the list — because a segment nobody can be recruited from cannot be sold to either.

Deliberately include people whose answers would hurt: those who tried something like this and stopped, those who evaluated an incumbent and stayed put, and those who have the problem and are content to live with it.

Success and failure criteria, both set now

Write both before any interview happens, and write them so a real outcome could fail.

  • Success: how many of how many participants, saying or doing what, unprompted. "Six of eight described the manual workaround before we mentioned it" is a criterion. "Participants confirm the pain is real" is not — every outcome satisfies it.
  • Failure: what result would falsify the thesis and send the decision the other way. This is the half people skip, and skipping it is what lets a disappointing round get reinterpreted afterwards into whatever the decision already wanted. If you cannot state a failure criterion, the plan cannot produce a "no" and there is no point running it.

Questions to avoid

Name the specific leading questions this thesis tempts you into — the ones you would drift toward at minute forty when a participant is being noncommittal and you want the decision confirmed. Write them out, in this thesis's own vocabulary, so an interviewer recognises the words in their own mouth. "Avoid leading questions" is not an entry; "Would you pay for something that did this automatically?" is.

Shape of the plan

Fit one conversation: roughly eight to twelve questions, ordered from broad context into specifics, with anything about money or switching near the end once trust exists. Leave room to follow what they raise — the plan is a spine, not a script.

Failure modes

  • A pitch wearing the costume of research.
  • Asking participants to design the product or estimate a market.
  • Hypothetical, future-tense questions: "would you", "could you see yourself".
  • Questions about frequency in the abstract rather than about the last occurrence.
  • Success criteria no realistic outcome could fail, or failure criteria left blank.
  • A plan that cannot produce a "no". If nothing the participants could say would change the decision, do not run it.

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.