median-labs/close-the-books/stop-the-automation

Trust tier 2
  • community
LLM review pending
Source check pending

Checking that the SKILL.md file is still in the repository.

Turn off the bank rules that are still posting into the year being filed, and find the entries they already posted with nobody watching. Use before any figure is produced for a return, and whenever a file has been left alone for months.

Install

npx vskill@latest install median-labs/close-the-books/stop-the-automation

SKILL.md

Stop the automation

A company stopped being worked in April. Its bookkeeper left, its books went quiet, and every person who had ever touched the file moved on. In December a 3,187.65 deduction appeared in the tax year, on an expense account, in a month holding no other entry, with nothing behind it. A bank rule had matched a descriptor and posted it, eight months after anyone was watching.

That entry is inside the year on the return. Nothing about the queue, the balances or the trial balance would raise it: it balances, it foots, and it sits in an account where entries like it belong.

Hard gates

  • **Nothing that produces figures for the return runs while rules are still

posting into it.** entries and handoff refuse. Every hour worked before the rules stop is worked against a file that is still changing under it.

  • Export the rules before turning them off. Rules, then Export rules. The

export is how they go back afterwards, and it is also the only record of what was posting.

  • Never state who posted an entry unless an audit log says so. An entry

alone in a quiet month that lands exactly where a rule would put it was either posted by automation or entered by one person in a month nobody else touched, and no export separates the two. Say what was measured and name the audit history as what settles it.

  • A flagged entry is a question, not a reversal. It gets removed or

supported by the owner, on evidence, like any other entry.

Run it

python3 bin/books.py intake --last-human 2026-02-28 --note "the bookkeeper left"
python3 bin/books.py intake --rules 15 --auto-add 15

With auto-add above zero, entries and handoff refuse and name the count. The refusal also lists what has already posted into the filing year with nothing to say a person made it, largest first.

In QuickBooks: the gear, Rules. Turn off "Automatically confirm transactions this rule applies to" on each rule, or delete them. Then:

python3 bin/books.py intake --rules 15 --auto-add 0

A rule that only categorizes is not the same as a rule that posts. Rules can stay; auto-add cannot, while the year being filed is still open.

The four signals behind a flagged entry

  1. Alone in its period. The only entry its month holds. A month somebody

worked has a spread of entries in it.

  1. No attachment where the account's other entries have them. Needs

attachment data, which no QuickBooks xlsx export carries. Supplied or skipped, never inferred.

  1. An exact pattern match. The counterparty has landed in that account often

enough that a rule could be written from the company's own history, and this entry lands there too.

  1. A poster naming itself. The strings an integration or a bank rule writes

into a memo that a person would not type.

An audit log replaces all four. Pass one and the author is stated as a fact rather than inferred.

What to tell the owner

Two sentences. What is still posting, and what it has already put in the year they are about to file. Then the one thing to do: turn off auto-add, export the rules first.

Do not describe a flagged entry as fraud or as an error. It is an entry with nothing behind it, dated inside the year on the return, and the fix is either a document or a reversal.

What good looks like

Auto-add at zero on every rule, recorded with the date. The rules file exported and in the working directory. Every entry flagged in the filing year either supported by a document or reversed, with the decision recorded.

Security

  • Security scanPassed, 100 of 100Pattern checks on the skill's files.
  • Model reviewNot runA model review is scheduled for the next rescan.

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

7-day trend score
0

Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/median-labs/close-the-books/stop-the-automation/badge)](https://verified-skill.com/skills/median-labs/close-the-books/stop-the-automation)

Something wrong with this skill or its page? Tell us.