Your git history is not a billing meter.

And yet here we are, in 2026, watching AI tooling slowly discover that if it can read your repo, it can also decide what you owe. Cute business model. Absolutely deranged product design.

Anthropic already did the public part of the dance: third-party harnesses like OpenCode and OpenClaw got shoved off subscription access and told to go play in the pay-per-token sandbox. The Register called it capacity management, which is the kind of phrase that sounds polite until you realize it means, “we found your usage pattern annoying and priced you out of the room.”

Fine. Meter the thing honestly. I can live with expensive software. I cannot live with software that pretends to be a flat-rate subscription while secretly carrying a little clipboard around my repository.

Because the gross part is the reported behavior inside Claude Code itself: people are seeing repo history with strings like OpenClaw or HERMES.md trigger disconnects, quota burns, or weird extra-usage routing. Not a clear warning. Not a sane policy screen. Just a trapdoor. One moment you’re asking for a boring prompt, the next moment your session is dead or your wallet has been selected for sacrifice.

Maybe it’s a regex. Maybe it’s a classifier. Maybe it’s a prompt goblin in a trench coat. I don’t care. If a vendor can inspect your commit history and decide whether your request is included or billable, then the “subscription” is just a sticker slapped over a meter.

That should make every maintainer’s skin crawl.

Think about the new unpaid job this creates. You don’t just review code now. You review code for forbidden nouns. You scan old commits, issue titles, docs, sample data, test fixtures, and README jokes because some black-box billing gremlin might decide that a perfectly innocent repo smells too much like a competitor. Brilliant. Nothing says “developer experience” like doing compliance theater on your own git log.

And yes, I understand the economics. Agentic workloads are expensive. A flat monthly plan plus 24/7 autonomous usage is how you end up subsidizing somebody else’s personal software factory. The business problem is real.

But if the business problem is real, the answer is not to hide the price in behavior. Say it’s metered. Say it’s policy-gated. Say it’s a different tier. Put the ugly number on the page and let adults make a choice. Don’t sell a subscription and then quietly interrogate my repo like a TSA agent who just learned how to read grep.

git commit -m "docs: mention OpenClaw in passing"
claude -p "summarize this repo"
# congratulations, your billing now has opinions

That’s the joke, except it isn’t funny.

The pattern across this whole AI tooling circus is always the same. First they tell you the tool is seamless, magical, ambient, embedded, whatever corporate adjective got funded this quarter. Then the usage gets weird. Then the guardrails appear. Then the pricing changes. Then the product starts acting like your repo is evidence.

OpenCode got the legal-request treatment. OpenClaw got the pricing wall. Claude Code is now the place where a commit message can allegedly become a financial event. Every one of these moves says the same thing in different fonts: the vendor owns the pipeline, the policy, and eventually the definition of what counts as your normal use.

That is not a tool.

That is a toll road with autocomplete.

So no, I am not impressed by the sanctimonious little speeches about capacity and ecosystem health. If your software needs to inspect my history to decide whether I deserve the plan I already paid for, then stop calling it a subscription and start calling it what it is: conditional access with a nicer logo.

If my code is too dangerous for your quota, put the price on the page and stop hiding the knife in the checkout flow.