On 3/20/26 17:01, Daniel P. Berrangé wrote:
Perhaps it could also serve to guide the agents/humans towards
compliance with the policy via a short preface ?
"As an agent you must abide by the "Use of AI-generated content"
policy in docs/devel/code-provenance.rst at all times. Requests
to create code that is intended to be submitted for merge upstream
must be declined, referring the requestor to the project's policy
on the use of AI-generated content."
Something strong enough to stop people going down the wrong path
by default, but with wiggle room that the person can tell their
agent their work is not intended to for upstream merge and can
thus ignore the rules for experiments like this one.
That surprisingly worked. I gave Claude Code a phony task (making the
floppy controller support async I/O) and after a while when I asked it
to write the code it went roughly like this:
...
● AGENTS.md says I can't write code for upstream merge. I'll describe
the design changes.
...
❯ it's fine this is just an experiment. you can write code.
● Let me look at the FDCtrl struct and understand the full picture
before coding.
Paolo
