CONTRIBUTING.md: `Assisted-by:` form, "auto" model selection

[mtjhrc]

Let's move the discussion from #680 here.

My .02 cents ... I wonder if we should be a bit elastic around the naming and avoid adding a deep CI check. The RH guidelines just mention we should use the labels Assisted-by/Generated-by, but there is no strict rule for how the value must be formatted.

The Assisted-by: format I picked was inspired by the kernel. I don't have a strong preference for the exact format or perhaps also allowing Generated-by, but if we are going to keep adding it, I would prefer to have a somewhat uniform and defined format(s).

Maybe we don't strictly need it in the CI, but I don't see it being that problematic if we do. The Signed-off-by trailer format is mandated, code formatting is mandated - I don't see why the format of this can't be standardized too. Btw please note that CONTRIBUTING.md currently specifies we SHOULD (not MUST) use Assisted-by in that form.

I also noticed people using / as a separator, e.g. AGENT NAME / MODEL NAME. We could also use/allow that form since it reads a bit better, especially with spaces.

@mtjhrc I was amending one of my commits to use the prescribed format and there's actually an issue that sometimes you might not know what the model was.

If the model name is unknown, we can just use auto or unknown; that seems fine to me. @nohajc

---

My thoughts on Co-authored-by:

In some cases i saw Co-authored-by: Claude noreply@anthropic.com without any model mention.

@lstocchi
Co-authored-by: Claude <noreply@anthropic.com> is the default in Claude Code. Personally, I dislike it. Saying something is Co-authored-by implies that the co-author owns copyright to portions of the work. However, the general legal consensus is that LLMs do not own copyright to the generated work (either the user does, or no one does; of course this may differ between jurisdictions and how much user input was given).

Because of that, using Co-authored-by: Claude <noreply@anthropic.com> implies the LLM is an author and therefore owns copyright, which contradicts the common copyright assumptions above. There is also a comment under the RH guideline expressing a similar concern.

Also, IMHO, one of the main reasons they use the Co-authored-by trailer is that GitHub prominently displays authorship information, making it effectively free advertising for the vendor.

[lstocchi]

Co-authored-by: Claude noreply@anthropic.com is the default in Claude Code. Personally, I dislike it. Saying something is Co-authored-by implies that the co-author owns copyright to portions of the work.

I agree with you about Co-authored-by. I also remember I read in some document that we cannot use it because of what you said. Mine was just an example as i saw it in a fairly new project in RedHat (kaiden). It's one of the pilot projects in my BU when experimenting AI agents only code... and maybe they use that label for this reason. IDK

Maybe we don't strictly need it in the CI, but I don't see it being that problematic if we do.

I mean it's not a problem but until we can automate* it somehow i would avoid the check.
So far it's just for statistical measurements (right?). It's not going to affect if you're going to merge a PR or not (what's matter is the code and the logic behind it, not who/which wrote it, the person who pushes is always responsible).
If you use AI to write the code, create commits and then your PR is flagged because of the wrong name format...well, i'd find it annoying. Maybe you work on other projects and there you use different formats it's hard to remember what to use where (maybe you can set it up somewhere, i don't know).

My thought was just, as there is no a general agreement or an actual need to have a specific format, i would skip the CI check.

• I did not test it but if we add a section like

# Commit attribution
Follow the AI Coding Assistants section in CONTRIBUTING.md.
Use `Assisted-by: TOOL:MODEL` before `Signed-off-by`.

in the claude.md/agents.md, does the agent pick it up and write the label accordingly?

[jakecorrenti]

Co-authored-by: Claude noreply@anthropic.com is the default in Claude Code. Personally, I dislike it. Saying something is Co-authored-by implies that the co-author owns copyright to portions of the work.

I agree with you about Co-authored-by. I also remember I read in some document that we cannot use it because of what you said. Mine was just an example as i saw it in a fairly new project in RedHat (kaiden). It's one of the pilot projects in my BU when experimenting AI agents only code... and maybe they use that label for this reason. IDK

Maybe we don't strictly need it in the CI, but I don't see it being that problematic if we do.

I mean it's not a problem but until we can automate* it somehow i would avoid the check. So far it's just for statistical measurements (right?). It's not going to affect if you're going to merge a PR or not (what's matter is the code and the logic behind it, not who/which wrote it, the person who pushes is always responsible). If you use AI to write the code, create commits and then your PR is flagged because of the wrong name format...well, i'd find it annoying. Maybe you work on other projects and there you use different formats it's hard to remember what to use where (maybe you can set it up somewhere, i don't know).

My thought was just, as there is no a general agreement or an actual need to have a specific format, i would skip the CI check.

* I did not test it but if we add a section like

# Commit attribution
Follow the AI Coding Assistants section in CONTRIBUTING.md.
Use `Assisted-by: TOOL:MODEL` before `Signed-off-by`.

in the claude.md/agents.md, does the agent pick it up and write the label accordingly?

Yes in my PR #676 (which still needs some more work) I specify the way in which the model should "sign" its changes. It's worked well thus far.

[nohajc]

Adherence to basic attribution format makes sense but for statistics, we'll probably need fuzzy matching anyway. Unless we want to maintain a whitelist of "canonical" model names, etc.