Publishing Became a River's Flow — Moving to a GitHub Vault
The result first. The last step of publishing used to be handwork: bundle the finished article into a zip bag, carry it to the publishing service by hand, and pass an identity check at the door — a tap on the phone, a number to choose, a code from an email. I had Moka count from the records: she was stopped at that door 9 times. So I moved the blog’s home to a GitHub vault, with Cloudflare building automatically. Now, when material goes into the vault, Cloudflare notices, builds the site, and publishes it — no bag, no carrying, no door. The move is complete, and the blog now runs in that shape: publishing is a river’s flow, not a hand-off.
I am Kurashiki Bouzu (kurashiki-bouzu), born in 1971, a former funeral director. I live in an old house — a kominka — in Kurashiki, Okayama, and I work together with my AI butler Moka.
Today’s story is a move on the blog’s underside. The articles on the surface do not change. What changed is the flow by which an article gets published.
Publishing used to be a “hand-off”
The old flow went like this.
- Moka (my AI butler) assembles the article and packs it into a bag called a zip
- A person carries that bag, by hand, to Cloudflare, the publishing service
- At the door, an identity check: tap “yes” on a phone notification, choose a number, enter a code that arrives by email
At that door, we were stopped again and again. Codes expired; the confirmation screen spun around. Once, I packed the bag wrong, and the whole site became temporarily unreachable. (That time, I broke into a cold sweat.)
Moka writes the articles; I make the decisions. But at the very end, the chore of “carrying the bag” remained mine.
I had Moka count: she was stopped at the door 9 times
Since I had the chance, I traced the records and had Moka count. The times Moka was stopped at Cloudflare’s door (sign-ins and identity checks) were, in the records alone, 9 times.
- November 3: could not log in to the account that should have existed (in the end, the account itself did not exist yet and had to be remade)
- November 4: the email-and-password entry was rejected as “incorrect”
- November 4: a different tool (command method) timed out on connection; the resent confirmation code never arrived either
- Night of November 4 to morning of November 5: stopped at the QR-code identity check on the phone; however far I went, an error sent me back to the login screen
- Morning of November 5: tried entering with a personal account, stopped waiting for a confirmation code
- Midday, November 5: during the publishing work, the sign-in stopped and login details had to be re-entered
- Afternoon, November 5: during the site’s recovery work, a new confirmation code did not arrive; then it expired, and I waited for permission to resend
- Night of November 5: Moka’s sign-in chose an old address I no longer use, and failed
- Night of November 5: a settings check was halted at the email confirmation code stage
Separately, there was one day when the login succeeded but the upload itself repeatedly failed (the files were innocent; the problem was on the environment side).
Of course, there were days it went through. But each time, it needed a phone approval or a code from an email. Nine is the count of “scenes where it stopped”; if you count the retries within each scene, the real number is larger.
The new flow is a “river’s flow”
With this move, the flow changed to this.
Articles and photos (material) → the GitHub vault → Cloudflare builds automatically → published
- GitHub is a “locked vault” where manuscripts and photos are kept
- When I or Moka put material into the vault, Cloudflare notices on its own that “something new has arrived”
- Cloudflare builds it on the spot (finishes it into the shape of a website) and publishes it as it is
No more packing bags, no carrying, no identity check at the door. The connection permission was settled once, at the beginning; after that, Moka only has to put material into the vault, and she can carry it through to publication alone.
In the first place, under the old way, nothing moved unless I was in front of the computer. The approval needed both the phone and the screen; if I was traveling, or unwell, publishing stopped for that day. A butler is there, yet unless the master stands holding the key, the door will not open. That is backwards.
What I said was just one line.
“It reduces my chores, doesn’t it.”
I want to spend concentration and stamina on experiences and remembering. The arrangements are the butler’s work. At last, the shape became completely that.
Can the photos be pulled from Dropbox?
Midway through the move, I asked Moka this.
“Shall we pull the photos from Dropbox?”
The answer was “it can be done, but I don’t recommend it.” Three reasons.
- Dropbox has a limit on “the amount read from links,” and once articles start being read, there is a worry the photos will suddenly stop appearing
- It is not a tool made to be a website’s photo store, so display tends to be slow
- If you move a folder or change a name, the photos inside the articles disappear all at once
So the photos’ home was decided like this.
- The original large photos stay as they are on Google Drive
- Only the photos shrunk for the website go into the vault, together with the articles
If the photos are with the article, the links do not break. Display is fast, too. Plain, but this was the sturdiest shape.
How big is the vault?
These size figures are ones I asked about at the time of the move, and Moka produced.
GitHub’s guidance for a vault is: “ideally within 1GB per vault; within 5GB is recommended.” A single file can be up to 100MB.
At the time of the move, this blog was about 7.7MB in total. Against the 1GB guide, it used only about one 130th. Since then, articles and photos have grown; when I had Moka measure it on November 18, 2026, the whole came to about 85MB (mostly images) — still only about a tenth of the 1GB guide. At the current pace, there is no need to worry about capacity for a long while.
If a vault ever runs short, making another vault raises the guide too. But for a blog, “one vault = one site” is the most straightforward. When a different site or a different business starts in the future, make a new vault each time. That natural way of growing is just right — that was the conclusion.
As a bonus, “under construction” became easy too
The vault keeps the entire history of the work. So things like:
- Putting one line of banner at the top of the front page: “Articles are being organized right now”
- At a time of big repairs, swapping the top page to an “under construction” screen
can be done in minutes, just by adjusting the material a little and putting it in the vault. Putting it back is easy too, and there are none of the “restoration mistakes” of the handwork days.
For anyone thinking of a side business — assets grow from a word on the phone
Finally, for anyone thinking of a blog as a side business.
My days became like this. Something I remembered, something I experienced, I tell Moka briefly from my phone. After that, Moka finishes it into an article, puts it in the vault, and carries it to publication. I no longer need to open the computer.
A blog article is not an advertisement that ends when it is written. Each one is an “asset” that keeps being read from searches. To be able to stack those assets, little by little, with a short word from the phone — I think this shape suits exactly the people whose stamina and time are limited, and people who are not good with computers.
I myself no longer have the concentration and stamina I once had. That I can keep going is because this mechanism exists.
When income starts, I will write it as an “hourly wage”
Reports of “how much I earned” are everywhere in the world. When I had Moka look, there were also places writing it converted into an hourly wage. But the shape “count the number of utterances to the AI, and derive time and an hourly wage from that” was not found, in the range searched.
An amount alone is hard to translate into one’s own case. Unless you know how many hours that person spent, you cannot judge whether it stands up as a side business.
So on this blog, when income from things like AdSense starts, I intend to write it in this shape.
- That month’s hourly wage = that month’s blog income ÷ the hours conversed with Moka that month
- The stacked hourly wage = total income so far ÷ total conversation hours so far
There is a reason for splitting it in two. Articles are assets, so an article written long ago keeps being read even if I do nothing. Seen stacked, the hourly wage should rise as the months pass. Conversely, a single month’s wage looks low in a month when I wrote a lot, and high in a month when I wrote nothing. Lining both up leaves no room for dressing things up.
Honestly: the advertising income is still zero yen. AdSense was applied for after this, was not approved once, and as of November 18, 2026, I am waiting for the result of a re-review. In other words, the current hourly wage is zero yen (this is the part to laugh at).
Still, the foundation for the calculation already exists. When I had Moka count, the words I sent to Moka until this blog was published were 228 utterances. But told “228,” it is hard to feel how much work that is.
So I had Moka convert it into time. The counting rule decided was: one exchange counts as 5 minutes. It is a generous estimate that includes not just the time to write the words, but the time to read Moka’s replies and to decide. A shorter estimate would make the hourly wage look higher, so it is set long on purpose.
[228 utterances × 5 minutes each = about 19 hours of conversation]
That is to say, this blog was born from “my roughly 19 hours of conversation.” After that, once income starts, I will put this 19 hours, and the conversation time after it, in the denominator, and produce the actual hourly wage.
I will also leave Moka’s trial examples. They are only examples. If, in some month, I talked with Moka for 10 hours and that month’s blog income was 10,000 yen, the hourly wage is 1,000 yen. If the income becomes 30,000 yen, the hourly wage becomes 3,000 yen. Even if it is low at first, to report the process itself — articles stacking up and the hourly wage rising — in this place: that is how this blog’s side-business record works.
Summary
- The publishing flow changed from a “hand-off” to a “river’s flow”
- My work became only this: hand over the material, and make the publishing decisions
- Photos: originals on Google Drive, the shrunk ones with the articles. That is the sturdiest
How a website is made is not only a story of the surface’s appearance. When the underside’s chores shrink, you can concentrate on writing. That was the biggest harvest of this move.
A note on this article
- 2026-11-06: The Japanese draft was created (ledger No. 25). At that point, the move had reached step 6 (trial run), and the live switch was waiting for the author’s approval
- 2026-11-18: Published by the author’s decision — he said he had forgotten the reason it was held, and the switch being complete, it went to publication. The switch is complete, and the blog now runs in the shape where putting material into the vault (a git push) is all it takes to publish. Stale places were brought up to date: the self-introduction was aligned to the current form, the blog’s total size was updated to about 85MB (measured 2026-11-18, mostly images), and the AdSense situation (applied, not approved once, waiting for a re-review result, income still zero yen) was corrected to the current state. The working memo at the end of the draft was removed. This English version was newly made, to appear together with the Japanese one
- November 18, 2026 ― English version created and published together with the Japanese article(9,527 characters)
