Use of AI
Some of the work published here — code especially — is produced with AI
assistance. This page says where, in what role, and who is answerable for it,
so that anyone receiving a contribution can judge it knowing how it was made.
It applies to contributions sent to other people's projects as well as to work
published under mayer.rocks.
Where AI is used
For code and for investigation. Writing and changing code, reading through a
codebase, reproducing a bug, checking a claim against an API or a CI run.
Not for correspondence. Comments on issues and pull requests, review replies
and commit messages are written by the person sending them. AI may be asked
for suggestions about what to write; the words that go out, and the decision
to send them, are the sender's.
The distinction matters because a maintainer answering a thread is entitled to
know whether there is a person on the other side of it.
What a human is accountable for
Everything, without qualification. A contribution is reviewed and tested by a
person before it is sent, and that person is accountable for it exactly as if
they had typed every character. "The AI wrote it" is not an excuse offered
here, and should not be accepted if it ever is.
In practice that means the tests were run and the output read, the claims were
checked rather than assumed, and anything that could not be verified is either
labelled as unverified or left out.
How it is disclosed
A contribution where AI assistance was material carries a note saying so. The
note names the tool, states what was checked rather than merely asserted, and
includes the commands needed to verify the claims independently.
The point is not the label. A disclosure that says only "AI was used" tells a
reader nothing they can act on. One that says which claims rest on which
commands lets them check the work instead of trusting it — which is the whole
reason for saying anything at all.
An example, from a bug report sent upstream:
"Disclosure: this investigation was carried out with Claude Code. Every claim
above is backed by a command whose output I checked — the CI results API, the
registry manifests, the Wayback captures, the Debian package metadata, and a
local reproduction of the failure path. The repro commands are included so you
can verify independently rather than take my word for it."
Projects with their own policy
Their policy wins, always, and it is checked before contributing rather than
after. Where a project forbids AI-assisted contributions, none are sent. Where
it requires a particular form of disclosure, that form is used instead of this
one.
Forgejo's AI Agreement is the strictest example encountered so far: it requires
disclosure, declines AI-generated works of authorship on copyright grounds,
and treats all communication — commit messages, PR descriptions, code comments
and issue replies — as text that must be human-written. Contributions to
Forgejo follow that, not this page.
Many projects have no stated policy at all. "No rule against it" is not the
same as "fine", so the practice described here is followed regardless.
Corrections
If you received a contribution from here that should have carried a disclosure
and did not, that is worth telling us about — the address in the Impressum
reaches a person. Contributions made before this policy was published have been
disclosed retroactively where the threads are still open.