agentleFS
andymaiandymai/claude-agent-team/skills/interview/SKILL.md

interview

Skill4 starsChanged yesterdayMITView on GitHub

Interview the user about a plan, a design, or a decision until each open decision is settled, before any plan or code. Use when the user says "grill me", "interview me", "stress-test this", or "poke holes in this", or asks if a plan is ready. Not for a clear task that is ready to do, and not for a session where no user can answer.

101 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use fetch-library-docs, contributor and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • Rehan-Ul-HaqSkill

    fetch-library-docs

    Fetches official documentation for external libraries and frameworks (React, Next.js, Prisma, FastAPI, Express, Tailwind, MongoDB, etc.) with 60-90% token savings via content-type filtering. Use this skill when implementing features using library APIs, debugging library-specific errors, troubleshooting configuration issues, installing or setting up frameworks, integrating third-party packages, upgrading between library versions, or looking up correct API patterns and best practices. Triggers automatically during coding work - fetch docs before writing library code to get correct patterns, not after guessing wrong.

    Also in 4 projects that use this

  • NimbleBrainIncSkill

    contributor

    Get started contributing to NimbleBrain open source — find an integration to build, set up your environment, propose an idea, or check on your work. Use when onboarding as a contributor. Triggers include "I'm a new contributor", "what should I build", "show me open issues", "file an issue".

    Also in 3 projects that use this

  • mjunaidcaSkill

    browsing-with-playwright

    Browser automation using Playwright MCP. Navigate websites, fill forms, click elements, take screenshots, and extract data. Use when tasks require web browsing, form submission, web scraping, UI testing, or any browser interaction. NOT when only fetching static content (use curl/wget instead).

    Also in 4 projects that use this

  • qa-amanSkill

    teach-me

    Teach a topic, video, article, codebase change, spec, or transcript until the user demonstrably understands it, using staged instruction, a running checklist doc, and quizzes. Use whenever the user says "teach me", "explain this in detail", "help me understand X", "/teach-me", "walk me through this", "I want to learn X", "understand this deeply", or pastes a YouTube URL, article, paper, PR, or spec and asks to understand it rather than summarize it. Also use when the user wants study notes or raw material from a source they are learning. Do NOT use for a quick factual lookup, a summary request, or when the user just wants the answer.

    Also in 8 projects that use this

  • trebaudSkill

    find-unknowns

    Workflow for surfacing a task's unknowns before, during, and after implementation — blind spot pass, brainstorm/prototype, interview, references, implementation plan, deviation notes, quiz. Use when starting work in an unfamiliar domain or codebase, when a task feels underspecified, when the user says "find my unknowns", "blindspot pass", "what am I missing", "help me scope this", or when a long task came back wrong.

    Also in 3 projects that use this

  • donatassimkusSkill

    offer

    Offer design, pricing strategy, and business model architecture. Invoke when asked about making an offer better, pricing a product or service, naming an offer, raising prices, generating fast cash, increasing lifetime value, finding the right customer, qualifying buyers, or designing a business model where customers pay for their own acquisition. Use when someone says their offer isn't converting, they need to charge more, they want to add bonuses or guarantees, or they need a money model for their business.

    Also in 3 projects that use this

The file

We split this skill into the rules it gives an agent

10 rules. 0 are common, 10 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 85 edited versions we read keep these.

  • 10

    Distinctive

    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:

    Write one sentence that says what you think the user wants.

    line 24
  2. Distinctive:

    Stop and wait for the answers.

    line 60
  3. Distinctive:

    After each round, record the answers, look up each fact that the answers make necessary, and find the new frontier.

    line 72
  4. Distinctive:

    When the frontier is empty, write the result.

    line 90
  5. Distinctive:

    Always fill "Out of scope".

    line 111
  6. Distinctive:

    No answer to the restatement, and then a new request: ask if the restatement is correct.

    line 120
  7. Distinctive:

    If the user corrects a line, change it and show the full restatement again.

    line 122
  8. Distinctive:

    After the yes, do these two things, and then end your turn:

    line 126
  9. Distinctive:

    Do not write a plan, a file, or code in the same turn.

    line 132
  10. Distinctive:

    Stop and say so when one of these conditions is true:

    line 136

“Common” is measured: kept by at least 6 in 10 of the 85 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

Interview

Interview the user until you and the user agree on what to build and on what not to build. You find the facts. The user makes the decisions. In this skill, you change no file and you start no work.

0. Check that a user can answer

This skill needs a user who answers in this session. If you run as a subagent, in CI, in a loop, or in a non-interactive run, do not start the interview. Report the open decisions as a blocker and stop.

1. State what you know

Write one sentence that says what you think the user wants. Then list which of these five items you cannot fill from the conversation:

  • User: who the result is for.
  • Reason: why the user wants it now.
  • Success: what a person can observe when it works.
  • Constraint: the limit that the result must not break (time, cost, compatibility).
  • Out of scope: what the user does not want built.

Show the sentence and the list to the user before the first question.

2. Separate facts from decisions

List every open point. Sort each one:

  • A fact has an answer in the environment: the repo, a config file, the output of a command, the docs. Find it yourself. Read the file, run the command, or start a subagent for a long search. Never ask the user for a fact that you can look up.
  • A decision has no answer until the user gives one. Put each decision to the user.

Gate: before you ask a question, you can say why no file and no command can answer it.

Example. The user says "add caching to the product page".

  • Fact: "Does the repo have a cache client?" Search the repo. Do not ask.
  • Decision: "Can a customer see a price that is five minutes old?" Only the user knows.

3. Ask in rounds

Some decisions depend on other decisions. The frontier is the set of decisions that depend on nothing that is still open. In each round:

  1. Take the decisions on the frontier. If there are more than four, take the four that the most other decisions depend on.
  2. Number each question. Give your recommended answer and one reason with it.
  3. Word each question so that "yes" accepts your recommendation.
  4. Stop and wait for the answers. Do not answer for the user.

Use this format:

Q1. <question title>: <the question, with the options if there are options>
Recommended: <your answer>, because <one reason>.

A recommendation that the user can reject is faster than an open question. The user corrects a wrong guess more easily than the user writes an answer from nothing.

After each round, record the answers, look up each fact that the answers make necessary, and find the new frontier. A question that depends on an open question belongs to a later round.

4. Test each answer

An answer that names a convention is not a decision. Watch for answers such as "scalable", "clean", "the standard way", "best practice", and "what most apps do". For each one, ask for the specific result:

  • "Scalable to how many requests each second, and by what date?"
  • "If you did not have to explain this choice to anyone, what do you want?"

An answer that gives the decision back to you ("whatever you think") is also not a decision. Give two concrete options and ask the user to select one.

5. Restate

When the frontier is empty, write the result. Use the user's own words where you can.

What I think you want:

- Outcome: <one line>
- User: <one line>
- Reason: <one line>
- Success: <one line that a person can observe>
- Constraint: <one line>
- Out of scope: <one line>

Decisions:
1. <decision>: <answer>
2. ...

Still open: <each point that the user chose to leave open, or "nothing">

Is this correct? Answer yes, or tell me which line to change.

Always fill "Out of scope". Many disagreements are about what the work does not include.

6. Get an explicit yes

The interview is complete only when the user confirms the restatement. These replies are not a yes:

  • "Sounds good" or "sure, go ahead": ask "Is there a line that you want to change?"
  • "Whatever you think is best": the user did not decide. Go back to step 4.
  • No answer to the restatement, and then a new request: ask if the restatement is correct.

If the user corrects a line, change it and show the full restatement again.

7. Stop

After the yes, do these two things, and then end your turn:

  1. Show the confirmed restatement one more time, as the final record.
  2. Tell the user the possible next steps: plan mode for an implementation plan, or /team:tickets to split the work into tickets.

Do not write a plan, a file, or code in the same turn. The user selects the next step.

Circuit breaker

Stop and say so when one of these conditions is true:

  • Five rounds are complete and the frontier is not empty.
  • Two rounds in a row end with more open decisions than they started with.

Tell the user the number of rounds, the decisions that stay open, and what you think is missing. Ask if the user wants to continue, to make the scope smaller, or to stop.

Final self-check

  • You asked no question that a file or a command can answer.
  • Each question had a recommended answer and a reason.
  • The restatement has all six lines, and "Out of scope" is not empty.
  • The user gave an explicit yes to the restatement.
  • You changed no file and started no work.

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.