Hi,

I think you got the point, I wasn't pushing to poison the agent.md in order to 
push back people in a negative way. I was sharing it because the idea of this 
kind of honeypot at the moment is the most effective that I saw in other 
projects. Than it can be tweaked in different way. But the core idea is that 
bots or vibe coders don't read the policy, and this is a way to tell 
immediately the mantainers that the user hasn't readed the code.

Regards, Ludovico

________________________________
Da: Tim Sutton <[email protected]>
Inviato: martedì, giugno 23, 2026 10:54:25 PM
A: Ludovico Toscano <[email protected]>
Cc: Alessandro Pasotti <[email protected]>; Nyall Dawson 
<[email protected]>; Vincent Picavet <[email protected]>; 
qgis-developer <[email protected]>
Oggetto: Re: [QGIS-Developer] R: Tightening AI submission policy?

Hi

On Tue, Jun 23, 2026 at 9:19 AM Ludovico Toscano via QGIS-Developer 
<[email protected]<mailto:[email protected]>> wrote:
Hi,

I'm a small contribuitor of qgis, it's my first time posting here and I was 
interest by this discussion.

I have an idea which is saw implemented in ghostty, to prevent contribution 
from bot or vibe coders. Basically the author of the repo has made a honeypot 
for these agents, he has made a AGENTS.md and a CLAUDE.md where in the last 
line of the document he tell the model to make a md file that saing "I am a 
sad, dumb little AI driver with no real skills." His idea is simple, if the 
author of the pr doesn't read the diff of the pr he will not notice that the 
LLM has added this file and this trigger a immediate ban.

https://github.com/ghostty-org/ghostty/blob/main/AGENTS.md

I hope this can be usefull,  given that a bot or vibe coder will not read the 
policy like he's not reading the code this solution can be a first filter.

I don't think degrading people is at all in line with our culture as a project. 
By all means put guidelines in place as to how AI can be used, but always deal 
with people in a positive way - for example the same mechanism could be used to 
add a line into the patch like "This line was automatically added by your LLM 
as a validation check. Please remove it before submitting your PR."

Regards

Tim



Regards, Ludovico.
________________________________
Da: QGIS-Developer 
<[email protected]<mailto:[email protected]>>
 per conto di Vincent Picavet via QGIS-Developer 
<[email protected]<mailto:[email protected]>>
Inviato: martedì, 23 giugno 2026 08:56
A: Alessandro Pasotti <[email protected]<mailto:[email protected]>>; Nyall 
Dawson <[email protected]<mailto:[email protected]>>
Cc: qgis-developer 
<[email protected]<mailto:[email protected]>>
Oggetto: Re: [QGIS-Developer] Tightening AI submission policy?


Hello,

Thanks for this proposal, I can only be in favor of a stricter policy.

I would also add a specific item on IP, even if I do not like pushing the 
burden to contributors while have no way of ensuring IP compliance.

In the line of :

- "By contributing to this project, you are responsible to ensure that you have 
the corresponding and sufficient intellectual property rights on all 
contributed code without exception to comply to this project's opensource 
licences and IP policies. Note that code generated by LLM has a high 
probability of copyright infringement."

Also a few resources for inspiration. I think if we can reuse other's positions 
word by word (if the idea is the same), it makes opensource project's position 
as a community stronger. Maybe we will end up converging towards a kind of 
standard opensource policy on AI, and that would be good for everyone ( and 
then one day the opensource world sues proprietary LLMs for copyright 
infringement, but that's another story).

-  NLnet; Policy on the use of Generative Artificial Intelligence for 
NLnet-funded projects : https://nlnet.nl/foundation/policies/generativeAI/
- Using LLMs Responsibly - Responsible AI Guide : https://responsibleai.guide/
- LLM-gen-AI - Software Freedom Conservancy : 
https://sfconservancy.org/llm-gen-ai/llm-backed-generative-ai-recommendations.html

Regards,

Vincent




On 23/06/2026 07:10, Alessandro Pasotti via QGIS-Developer wrote:

Big +1

It's probably hard to put in clear terms (and forgive my English), but
I think that the criteria should be even stricter than "contribute
only what you understand":

"Do not contribute anything that you wouldn't have been fully capable
of writing on your own without any AI assistance."

or

"Contribute only what you could have done by yourself without any AI
assistance."

Tthis makes it more likely that you will be able to fix or improve
your work if needed.

On Tue, Jun 23, 2026 at 1:47 AM Nyall Dawson via QGIS-Developer
<[email protected]><mailto:[email protected]> wrote:


Hi lists,

Following recent discussions and conflict within the PostGIS community, I would 
like us to consider tightening our AI submission policy to hopefully avoid 
similar conflict arising in our community.

I would propose that we move away from our current (quite permissive) "human in 
the loop" policy (see 
https://github.com/qgis/QGIS-Enhancement-Proposals/blob/master/qep-408-ai-tool-policy.md
 ) to something more restrictive, such as what the GDAL project has done. See 
https://github.com/OSGeo/gdal/blob/33399bf734f48a6727050ccd13a43575d43f6bf4/doc/source/community/ai_tool_policy.rst).

In short, GDAL's policy is:

Contributors can make limited use of LLMs for contributions in GDAL, subject to 
details mentioned below:

- Human contributors must be the primary author(s) of GDAL contributions
- All contributions including code, ticket comments, and commit messages should 
be fully understood by the author(s) submitting them to the project.
- Submission of vibe-coded contributions is banned.
- LLMs may only be used as an improved auto-completion mechanism, or for 
repeated tasks (mechanical refactoring) that could potentially be completed 
with a deterministic algorithm.
- Human-coordinated or uncoordinated (OpenClaw, etc) use of agents for 
submission of contributions to the GDAL repository is banned.
- Any LLM usage must be indicated by ticket label, comment, or commit message 
indication and account for what was written by whom/what.
- The contributing human author is ultimately responsible for every line of 
code, comment, or mailing list interaction they initiate, and all of it is 
subject to the project's :ref:`code_of_conduct`.
- The typical high verbosity of LLM code and text is actively discouraged. More 
code is more code to maintain. High verbosity contribution (tickets, code, 
messages, etc) will be seen as indication of LLM-generated content when not 
labeled otherwise and may be ignored, closed, left unmerged, or removed at 
maintainers' discretion.


I personally think a combination of GDAL policy + borrowing the "contribute 
only what you understand" term from Godot's current policy (see 
https://contributing.godotengine.org/en/latest/pull_requests/pull_request_guidelines.html#ai-assisted-contributions
 could be a good step forward. Specifically this clause:

Contribute only what you understand
Only submit code that you fully understand and are prepared to explain to a 
maintainer. This especially applies if you implement the idea of another 
person, copy code from elsewhere, or if you use AI to assist you with your 
contribution. In all of these cases, you must disclose which part of your 
submission wasn't fully authored by you.

Thoughts?

Nyall


_______________________________________________
QGIS-Developer mailing list
[email protected]<mailto:[email protected]>
List info: https://lists.osgeo.org/mailman/listinfo/qgis-developer
Unsubscribe: https://lists.osgeo.org/mailman/listinfo/qgis-developer


_______________________________________________
QGIS-Developer mailing list
[email protected]<mailto:[email protected]>
List info: https://lists.osgeo.org/mailman/listinfo/qgis-developer
Unsubscribe: https://lists.osgeo.org/mailman/listinfo/qgis-developer


--

Tim Sutton
Kartoza Cofounder
Tim is a member of the QGIS Project Steering Committee

T : +27(0) 87 809 2702          E : [email protected]<mailto:[email protected]>   
       W : kartoza.com<http://kartoza.com/>

[https://kartoza.com/files/KartozaEmailSignature.gif]

This email and any attachments are confidential and intended solely for the use 
of the individual or entity to whom they are addressed. If you
have received this email in error, please notify the sender immediately and 
delete it from your system. Unauthorised use, disclosure, or copying
of the contents is prohibited.

_______________________________________________
QGIS-Developer mailing list
[email protected]
List info: https://lists.osgeo.org/mailman/listinfo/qgis-developer
Unsubscribe: https://lists.osgeo.org/mailman/listinfo/qgis-developer

Reply via email to