1 The client
Hi! We open at 9 on Saturdays now, not 10. Also the deluxe cut is $45.
Got it — I’ll send you a preview shortly.
Running live on two client sites today
One label.
One merge.The site is live.
A small-business client texts you a change. It lands as an issue, an AI agent edits the content and opens a pull request with a preview link, and you tap merge from your phone. That’s the whole job.
$79 one-time · lifetime updates
Online guide, updated as the system changes Offline copy + starter code (MIT) No subscription
client texts or emails
issue
you: one label
agent opens a PR
you: merge
live
Six stages. You touch exactly two of them.
A real change, start to finish
Saturday hours changed. Here is every step you actually do.
Nothing below is mocked up for a sales page. This is the shape of the system that runs two live client sites — and it is the same shape whether you have two clients or twenty.
2 You: one label
#218 Saturday hours + deluxe price
agent:applyopen
3 The agent
Update Saturday hours & deluxe price
+1−1 content/hours.yml
+1−1 content/services.yml
preview ready deploy-preview-218
4 You: merge
merged #219
Live · 54 seconds later
Reply sent to the client.
The arithmetic
The same request, with and without the pipeline.
Care plans die because a ten-minute edit costs forty minutes of context switching. This is the part the pipeline actually fixes.
By hand
The way most people maintain a client site
- Read the message. Remember which client, which repo, which branch.
- Open the laptop. Pull, install, start the dev server.
- Hunt for where hours live. Was it YAML? The CMS? Hard-coded?
- Edit, check it locally, remember the Spanish version too.
- Commit, push, wait for the build, load the site to confirm.
- Write the client back. Hope you did not forget anything.
Cost: a laptop, a context switch, and most of an afternoon’s focus.
With the pipeline
What you do instead
- The message becomes an issue in the client’s repository.
- You add one label.
- You open the preview link on your phone and look at it.
- You tap merge.
Two actions. No laptop required.
What is in the guide
Written from the system that is actually running.
Not a theory of how this could work. The architecture, the prompts, the routine, and a dated list of everything that has broken so far.
01
The architecture
Every piece of the stack and why it beat the alternative that looked obvious. The decisions are shown with their reasoning, so you can swap a piece out when your situation differs.
02
Setup, as exact prompts
Every setup step written as the literal prompt used to get it done — with full manual instructions underneath, in case your agent does something different.
03
The daily routine
What running this looks like once it is up: triage, labels, previews, merges, and how to keep a client feeling attended to without opening a terminal.
04
The failure catalog
Every way the system has broken so far, dated, each with the fix that worked. The chapter most guides leave out because it requires actually shipping the thing.
05
The capacity model
Real measurements with their sample sizes, so you can plug in your own numbers instead of trusting a round figure someone invented for a pitch deck.
06
The starter code
The workflows and config as a zip you can drop into a repository, MIT licensed. Read it, fork it, or rewrite it — the guide explains what each part is for.
Before you buy
Honest about who this is for.
Buy it if
- You are comfortable in a terminal and can read a YAML file.
- You maintain — or want to maintain — sites for small local businesses.
- You already have a client or two and the edits are eating your week.
- You would rather adopt a system that runs than design one from scratch.
- You want the failure modes handed to you instead of discovering them live.
Skip it if
- You want a course on finding clients or selling care plans. You bring the clients.
- You are looking for a hosted product with a dashboard and a login.
- You have never opened a terminal and do not want to start now.
- You need something that works with no GitHub repository involved.
You do not need to be a working developer. You do need to be willing to read a config file.
What lands in your inbox
Three things, one payment, no renewal.
Primary
The online guide
Lifetime access, and it keeps getting updated as the system changes. Bookmark it and re-read the failure catalog when something breaks.
web · lifetime
Backup
An offline copy
The whole guide as a zip that opens in any browser. Yours to keep, no matter what happens to any website later.
.zip · opens anywhere
Code
The starter repo
The workflows and configuration that make the pipeline run, as a zip. MIT licensed — use it in client work without asking.
.zip · MIT
Questions
The things people ask first.
The guide documents the exact setup that runs today, including the agent and the rules file it reads before touching content. The architecture chapter explains why that choice beat the alternatives, so swapping in a different agent is a documented change rather than a rewrite.
No. The agent only ever opens a pull request with a preview link. Nothing reaches the live site until you look at the preview and merge. That human gate is the point of the design, not an afterthought bolted on.
You see it in the preview and you do not merge. The failure catalog is a dated list of every way this has gone wrong so far, each with the fix — including the ones where the edit was technically correct and still wrong for the client.
It works for any site whose content lives in a repository you control and that deploys on push with preview builds. The setup chapter covers getting an existing site into that shape, which is usually the real work.
The capacity chapter gives real measurements with their sample sizes rather than a confident round number. Your answer depends on how chatty your clients are, and the model is built so you can substitute your own figures.
No. One payment of $79 buys lifetime access to the online guide including future updates, plus the offline copy and the starter code as downloads you keep.
A small-business client texts or emails a change: new hours, a new price, a photo of the new sign. It becomes an issue in their repository. You add one label. An AI agent reads the site's rules, edits the content, and opens a pull request with a preview link and a reply drafted in the client's language. You check the preview on your phone and merge. That's the whole job: one label, one merge.
This guide is that system, written from the one I actually run:
- The architecture, and why each piece beat the alternative
- Every setup step as the exact prompt used, with full manual instructions underneath if your agent does something different
- The daily routine once it's running
- A catalog of every way it has broken so far, dated, with the fix
- A capacity model with the real measurements and sample sizes, so you can plug in your own numbers
Who it's for: someone comfortable in a terminal who can read a YAML file. You don't need to be a working developer.
What it isn't: a course on finding clients or selling care plans. You bring the clients.
What you get:
- Lifetime access to the online guide, which keeps getting updated
- An offline copy (zip) that opens in any browser
- The starter code (zip), MIT licensed
Stop opening the laptop for a price change.
The Website Maintenance Pipeline — lifetime access to the guide, the offline copy, and the starter code. One payment.
Instant delivery. The online guide is updated as the system changes, and your access does not expire.