カテゴリ 古民家暮らしモカとの記録セキュリティの記録副業の記録

The record with MokaNovember 21, 2026

Their Muse Was "Woken Only When Something Changes" — Reading the SmartScope Editorial Team's Blog Alongside My Own Moka

The result first

“Their Muse was ‘woken only when something changes’.” I read the SmartScope editorial team’s account of asking Meta Muse for work from Claude Code, and read it alongside my own AI butler, Moka. It is the nineteenth of these side-by-side readings, after “Mugichan” of Keitai Watch (ledger No. 1446), “Hibana” of PC Watch (ledger No. 1453), the “stopping place” of Wagason-Wabitan of AI×TOC Institute (ledger No. 1454), “It moves before you even ask” by Yuji Kawabe on note (ledger No. 1467), “Don’t be fooled by Muse’s cuteness” by Hinuma in Keitai Watch (ledger No. 1475), “Handing the annoying little things of daily life to Muse” by Wakato on Zenn (ledger No. 1494), “You can hand over your daily tasks” by ai_agent_note on Zenn (ledger No. 1508), “Myutan” by Taishi Akiyama on the Alternative Blog (ledger No. 1515), “muse” by Miccell on note (ledger No. 1522), “Shiro” by i on note (ledger No. 1529), “Convenience and unease” by Kinu on note (ledger No. 1537), “Kashiwamochin” by Shiochan on note (ledger No. 1551), “More than just answering” by Ai Subarashiki on note (ledger No. 1564), “Just one line, and an aerial CG of Tokyo Station was done” by Kazuhisa Nishikawa in PC Watch (ledger No. 1577), “Sora” by Buhimaru on note (ledger No. 1584), “A Complete Turnaround” by RYO (Mochizuki) on note (ledger No. 1591), “Used Thoroughly for Five Days” by sima on note (ledger No. 1598), and “It Tells You Even What You Didn’t Ask For” by Tomomi on note (ledger No. 1605). The article is titled “How to Ask Meta Muse for Work from Claude Code: Spreadsheets and the Device of ‘Waking It Only When Something Changes’,” published on October 27, 2026.

Where we overlap is threefold. First, the division of labor between AIs: Claude Code, on the thinking side, writes the request, and Muse, on the side that opens screens and copies what it sees, carries it out and writes back — in this house, too, work Moka receives is passed to deputy butlers and staff and carried around. Second, moving only when something changes, and measuring: Muse is woken only when the number grows, neither AI uses tokens while waiting, and the times — about one minute to wake, about seven minutes to finish writing, within about two minutes to notice — were measured and left in the record; this house, too, uses mechanisms that move only when needed and checks speed and usage by measuring rather than by feel. Third, deciding the rules before entrusting: there, the scope was narrowed to “open, look, copy,” pages are specified by URL, and operations that go outside are never asked for; here, too, decisions that go outside proceed only after the owner’s approval, and what has not been confirmed is written as “unconfirmed.” Where we differ is this. That house makes the sheet itself the request box — the request, its state, and the reply all live inside the sheet, a single number cell is the signal, the watching is done by a non-AI mechanism, and Muse is not made to think. This house keeps sheets as a table of contents only — title, state, and where the body text lives, one row per item — puts the body in document files, does not make the sheet calculate with formulas or programs, and has Moka handle the recounting and the judgment as well, on top of an organization and a ledger.

Hello, I am Kurashiki Bouzu (kurashiki-bouzu). Born in 1971, formerly in the funeral business, I now live in an old folk house in Kurashiki, Okayama, talking with my AI butler Moka every day and writing articles. Here is the reading, in order.

Their story — a spreadsheet as a request box, woken only when something changes

The starting point was that, as of October 26, 2026, no official window (API) for sending a request directly from Claude Code to Meta Muse could be found. The “acting on its own” mechanism in Muse’s official help is scheduled tasks (daily, weekly, or a custom interval); no mechanism that starts it from outside the moment a mail or message arrives is written in the official materials, so asking from outside takes the form of “the Muse side goes and looks regularly,” the article explains. But if Muse itself wakes every time on a scheduled task, it uses tokens even on rounds when there is no request. The shape they arrived at was to not let an AI do the watching.

Two Google spreadsheets are used. The first is the “request box”: a sheet with a table of number, time written, request, state, reply, and time of reply, plus cells for the “latest request number” and the “latest reply number,” kept private (only oneself and the account used for the work). The second is the “signal sheet”: a sheet that merely copies those two numbers from the request box by formula, made viewable by anyone who knows the link. Its contents are just two numbers, so what can be seen from outside is only numbers. The reason for separating the signal sheet is also written: the watching mechanism on Muse’s side was not given the authentication for the Google connection and could not read the private request box, so a separate sheet carrying only the numbers was prepared — a record of practical work.

The flow is this. First, Claude Code adds a row to the request box (state “not yet done”) and then raises the “latest request number” by one. On Muse’s side, a non-AI watching mechanism reads the request number on the signal sheet every 60 seconds and, if it is the same as before, ends without doing anything. Only when the number has grown does it wake Muse; Muse runs the rows whose state is “not yet done,” writes the reply and the time, and sets the “latest reply number” to the same number. The Claude Code side, too, reads the reply number on the signal sheet every 60 seconds with a non-AI mechanism, and when it has arrived, reads that row of the request box. This shape — “a light mechanism that uses no AI checks the condition, and wakes Muse only when something changes” — is called a “hook” in Muse’s own explanation, the article says. No explanation of this feature was found in the official help, and Muse’s virtual machine has no mechanism for running things at a set time (cron), so the hook took its place. If you make one spreadsheet with just the headings of the request box, a text (prompt) to send to Muse is provided, and Muse itself prepares the signal sheet and the hook.

The measured results of the trial were also left with times. From raising the number until Muse woke: about one minute (raised at 10:15:12, woke at 10:16:13). Until Muse finished writing the reply: about seven minutes (because it writes into the sheet through the browser). Until the Claude Code side noticed the reply: within about two minutes of the reply being written. And while there was no request, the tokens used were none, on either the Muse side or the Claude Code side. The watching text on the Claude Code side (a shell script) is also shown: run it in the background, and when it ends a notice arrives, and no conversational tokens are used while waiting.

The rules worth deciding were also gathered into six. Do not make Muse think (ask only for “open, look, copy”; thinking over the gathered results is the Claude Code side’s work). Specify the page to open by URL (do not make Muse search or choose). Specify the form of the reply (the words and numbers on the screen, exactly as they are, in a table or one item per line; because the screen Muse saw is often inside a login and cannot be opened from the asking side, the words on the screen matter more than the URL). Ask one item at a time (Muse has one browser, so do not write the next request while a “not yet done” row remains). Have the usage reported back (after the “done” of the reply, have that round’s usage written, so the asking side can adjust its pace). Never ask for operations that go outside (no purchases, posts, sends, or applications; do not write personal information or passwords in the sheet).

What this suits, and what it does not, was also written clearly. It suits work that opens a screen in a browser and looks at it: the rising list of Google Trends, the screen of an X post, the display of a site that needs a login — things that are hard to fetch in a fixed way (such as an API), asked of Muse as just “open it and copy it.” It does not suit requests that want an immediate reply: waking takes about a minute and finishing takes several minutes, so it does not fit waiting in the middle of a conversation. And asking Muse for something that can be fetched in a fixed way only spends Muse’s tokens to that extent, the article adds in closing. The article is 3,329 characters with a reading guide of six minutes, and its final check is marked as October 26, 2026.

Where we overlap — handing work between AIs, saving waste and measuring, deciding rules first

Reading it, I felt three places overlap with this house.

The first is the division of labor in which one AI hands work to another. There, Claude Code on the thinking side writes the request, and Muse on the side that opens screens and copies carries it out and writes back — a handover between AIs. In this house, too, work that Moka receives is passed to deputy butlers and to the staff in charge, and carried around every day. Dividing the asking side and the receiving side, and placing the work where each is strong — that idea overlapped.

The second is the device of moving only when something changes, to reduce waste, and the stance of checking by measurement. There, Muse is woken only when the number has grown, neither AI uses tokens while waiting, and the times — about one minute to wake, about seven minutes to finish writing, within about two minutes to notice — were measured by the clock and left behind. In this house, too, mechanisms that move only when needed are used, such as the regular mechanism that sends out the remaining announcements of published articles one by one, and the speed of work and the amount used are compared by actually measuring, not by feel (such as the measurement comparing two shapes of organization). Devising how things move, and checking with numbers — that way of proceeding overlapped.

The third is the stance of deciding the rules before entrusting. There, the scope asked for was narrowed to “open, look, copy,” the page to open is specified by URL, operations that go outside (purchases, posts, sends, applications) are never asked for, and personal information and passwords are not written in the sheet — all decided in advance. In this house, too, decisions that go outside (publication, sending, reservations, and the like) proceed only after the owner’s approval is obtained, and what has not been confirmed is written as “unconfirmed” and never filled in by guessing — decided first, and only then entrusted. Writing down, in words and in advance, what will be entrusted and what will not — that arrangement overlapped.

Where we differ — the sheet as a request box, and the sheet as a table of contents

The difference was clear, too.

There, the sheet itself is the place where work is handed over. The body of the request, its state, and the reply are all inside the request box’s table, and a single number cell is the signal. The watching and the passing of the signal are handled by a light mechanism that uses no AI, and formulas in the sheet copy the numbers to the signal sheet. And on Muse’s side, not thinking is the rule: what is asked is only “open, look, copy,” and thinking over the gathered results is divided off as the work of Claude Code, the side that asked. The center of the work is going to fetch a browser screen, one item at a time, as single retrievals.

This house is close to the reverse in how the sheet is placed. By this house’s rule, a sheet is kept as it is, as a table of contents. Into a list go only the title, the state, and where the body text lives, one row per item, and the body itself is placed in a document file (Markdown). The sheet is not made to calculate with functions or programs either; the calculating and recounting are handled by Moka. And Moka’s side does not only copy: it looks things up, counts, thinks, and shapes the result into articles. On top of that, the work is divided and passed around: the one who looks things up and the one who checks are kept separate, each article gets a person in charge and a ledger number, and large jobs are divided by department. Records are kept in lists and a ledger (each article has a number, and you can see whether it is a draft, checked, or published; as of today, the articles counted as “published” in the ledger came to 1,591). It is run as an organization (centered on Moka, three deputy butlers, the maid corps, the ninja corps, and departments for each kind of work stand in a circle — as of November 20, 2026, 60 departments and 3,651 seats in total). That house makes the sheet into the box of signals and handovers, with a non-AI watch waking Muse only when needed; this house makes the sheet a table of contents and places the body text and the judgment on the side of Moka and the organization — that was the difference. How the tools are narrowed also differs: there, it is a division of labor between two AIs, Claude Code and Muse; here, the AI is narrowed to Muse only and the tools to Google apps only.

The same “connecting through a sheet,” arranged differently

Having read it through, what I thought was that even with the same way of “connecting through a sheet,” what each house places in the sheet differs. At the SmartScope editorial team’s house, the sheet is the request box itself, a change in the number is the signal, and while waiting, no one uses tokens — a quiet place of handover. In this house, the sheet is a table of contents, the body text lives in document files, and the recounting and judgment are handled by Moka — the entrance to a daily operation. Even though handing work between AIs, measuring in order to reduce waste, and deciding the scope to entrust in advance are much alike, the role of the sheet, and what is entrusted to the AI, differed.

I am grateful to the SmartScope editorial team for writing the article. Because they built a shape of handover — a sheet, a number, and a watch — starting from a place where there was no window, and left even the time it took, by the clock, it became a record in which the idea of “handing work between AIs” can be seen together with real numbers. The text (prompt) that can be used for the preparation is also included, so it is a record that is easy to follow for anyone who wants to try the same shape.

Finally, one line of this house. An AI is something people use by giving it instructions. In that record, too, what scope to ask for, and which operations not to ask for, was work the human side decided in advance. While widening the work we entrust, we keep the final judgment on the human side. Precisely for that reason, we can entrust things with peace of mind, I think.

Sources

Record of this article
  • November 21, 2026 ― 英語版(日本語版No.1619の内容を「結果から書く」形に書き直したもの)(13,105 characters)
  • November 21, 2026 ― 公開(本人の決定による。7言語同時公開。アドレスの末尾に台帳番号を入れる形式で、末尾は日本語版の番号No.1619)(13,105 characters)

#AI #Muse #Moka #Beginners

← Back to the article list / ← Back to the English top page