agentleFS
median-labsmedian-labs/close-the-books/skills/ask-the-founder/SKILL.md

ask-the-founder

Skill1 starsChanged 2 months agoMITView on GitHub

Turn the things the books cannot settle into a short, specific list of questions for the owner, and record their answers with a date and a source. Use whenever a categorization, an entry or a treatment is blocked on something only the owner knows.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    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:

    Every question specific enough to answer from memory.

    line 92

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 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

Ask the founder

The scarcest resource in a catch-up is the owner's attention. Everything here is about spending it well.

A list of thirty questions gets zero answers. Five specific ones get five answers, and usually a sixth thing you did not know to ask.

Five questions that are not in this round

The bank's own balance, the For Review count, the feed state, the reconciled date and the bank rules come off a QuickBooks screen rather than out of memory, so they have their own short round with the screen named against each one. See read-the-screen. Asking them here mixes a lookup in with questions that need the owner to remember something, and the lookup is the half that gets skipped.

The other list of questions

What a return needs is a separate list, raised out of the chart rather than out of a stuck transaction, and ready-to-file builds it. Those questions live in the same store and come out in the same rounds, so books.py questions batches both together and books.py answer records both the same way. The difference is where they come from: this skill asks what a categorization is blocked on, and that one asks what the firm filing the return will otherwise chase in April.

Hard gates

  • Never ask what the books already answer. Mine the history first. Asking where a vendor goes when the last forty transactions all went to the same place wastes the one thing you are short of and makes the tool look like it is not paying attention.
  • One question per paragraph, each ending in a question mark. An implied question gets an implied answer.
  • Never ask for a decision you should be making. "Should this be an expense or an asset?" is usually your job. "Was this trip for the Berlin office or personal?" is theirs. The test: does answering it require knowing something that happened in the world, or knowing accounting?
  • Record every answer with a date and a source. An undated answer is indistinguishable from a guess in a month.
  • Ten at a time, maximum, ordered by what unblocks the most work.
  • Never answer one on their behalf, and never leave one blank. Where they cannot answer, record that they could not and why: books.py answer Q03 --cannot "the invoice is with our former bookkeeper". A stated unknown can be chased. A blank reads downstream as a no.

Run it

python3 bin/books.py questions --profile profiles/mine.local.json --round 1

Writes questions/round-1.md, drawn from what the profile could not settle and what the categorizer parked. Each question carries the vendor, the amount, the date and, where it helps, what you will do with each possible answer, which often lets them answer in one word.

Record answers as they come:

python3 bin/books.py answer Q03 "Berlin office, business trip" --on 2026-09-12

Round two does not open until round one is answered. Two open rounds means neither gets finished.

What a good question looks like

Not: "Can you clarify some uncategorized transactions?"

But: "There's a 4,200 payment to Fabrikam Freight on 14 March that doesn't match anything else that year. Was that a one-off shipment, or the start of something recurring?"

The second one can be answered from memory in ten seconds, standing up, on a phone. That is the bar.

Questions that come up in almost every catch-up

Worth asking early, because each one unblocks a category of work: who was keeping the books and when did they stop; which of these accounts are still open; are there accounts not in QuickBooks at all; which card is the real one where two look alike; are any of these payments personal; did anything unusual happen this year, such as a raise, a loan, a sale, or a new entity.

What good looks like

Every question specific enough to answer from memory. Every answer stored with its date and who gave it. Nothing asked twice. And the count going down each round, which is the only real measure that this is converging.

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.