How it works
How the work actually happens.
You’re being asked to let one person and a governed toolset near a production container. “Trust me” isn’t an answer, so here’s the mechanism. Every guarantee below is a property of the code, not a promise about my intentions.
The mechanism
Five steps, and no dashboard in any of them.
There is no console to log into, because the interface is a conversation and the output is a file. Here is the whole path a change takes.
-
It starts with a sentence — usually yours.
- Audit this container.
- Add the LinkedIn conversion tag.
- Did checkout events drop last week?
You send it in an email; I run it. No syntax to get right, no permissions dance, no ticket.
-
Claude drives the tools.
Claude Code holds Project5’s governed toolset and picks the ones the job needs — reading the container, writing the entity, publishing the version, reading GA4 back. Every write lands in a GTM workspace first: nothing reaches the live container until a version is created and published as a separate, deliberate step.
-
Every change is versioned.
Mutations land as container versions with a journal entry attached, and the container’s full contents are committed as files, so any version can be diffed against any other. GTM answers “what changed, and when” too — inside its own console, one container at a time, for as long as you hold a login. This answers it as text you own, outside Google, and it keeps answering after your access ends.
-
Rollback is a sentence.
Roll back to the version before that. The publish reverses, and the reversal is journaled too. That insurance is what makes delegating the container safe in the first place.
-
Writes are fenced.
An allowlist names the containers automation may touch. Anything outside it fails closed — not a warning to click past, a refusal. Read access and write access are not the same grant.
Straight answers
Who is actually running this.
- There is no self-serve version
- No signup, no installer, no seat to buy. Everything on the price list is delivered by me running the tool and sending you the result. If a product you can log into is what you are shopping for, I am the wrong vendor and it is cheaper for both of us to know now.
- The write path is fenced whoever is at the keyboard
- The allowlist is a check in the code that runs before any write call and refuses anything not named on it. It is empty unless it is deliberately set, so the default state of the software is that it cannot write to anything at all — including for me.
- An audit is read-only, and read-only is enforced by Google
- The free audit asks for GTM’s Read permission on your container. That role carries no write capability, which is Google’s enforcement rather than mine — and it is revocable from the same screen you granted it on, with no notice to me.
- If you run Claude Code yourself, say so
- There is a real second audience here: teams who would rather hold the read-only tools themselves. That is not something you can buy today. Tell me in the request note anyway — it is a conversation I want to be having.
Read-only. Name a container, get the scored report back.