contrib : rewrite AGENTS.md, make it more clear about project values#21270
Conversation
| The following will result in immediate PR closure: | ||
|
|
||
| For related documentation on building, testing, and guidelines, please refer to: | ||
| - **AI-written PR descriptions or commit messages** - these are typically recognizable and waste reviewer time | ||
| - **AI-generated responses to reviewer comments** - this undermines the human-to-human interaction fundamental to code review | ||
| - **Implementing features without understanding the codebase** - particularly new model support or architectural changes | ||
| - **Automated commits or PR submissions** - this may spam maintainers and can result in contributor bans |
There was a problem hiding this comment.
Writing code is no longer forbidden here. Note that I initially wrote this point as "no wall-of-text" and the AI rephrased it such that it sounds more professional
There was a problem hiding this comment.
AGENTS.md is loaded as system instructions by agents. The "Guidelines for Contributors" section is project policy for humans that fills agent context without making them more effective (that's peanuts, but still...). Should that part stay in CONTRIBUTING.md, keeping AGENTS.md focused on what agents should/shouldn't do? 🤔
There was a problem hiding this comment.
That is a good point. It may be good to keep it shorter and only address the agent itself.
There was a problem hiding this comment.
The current content takes around ~1200 tokens which is I think is still somewhat acceptable. Ofc it is possible to reduce it to around 500 tokens (will be a follow up PR). I don't think a second file will work, because in most cases there are no reasons for the agent to read it
There was a problem hiding this comment.
It's obviously possible to create a prompt that formats everything perfectly down to the smallest detail, like a human hardcore programmer, but it would be even harder to check. What a paradox
There was a problem hiding this comment.
Why not simply give the assistant a behavior that explains to the user that they need to know precisely what they want to do, that they must ask all the necessary architectural questions, and that the work of others must be respected, and that humans have to proofread everything and that it takes time? The perfect Inception approach…
There was a problem hiding this comment.
IMO prompting to change the behavior of the AI is almost like role play, and quite tricky to get right with heavily guarded models like claude or gpt.
Anw, I'm merging this PR as-is because it should already unblock some legitimate use cases for maintainers / contributors. Any improvement to this can be done via follow-up PRs.
…gml-org#21270) * contrib : rewrite AGENTS.md, make it more clear about types of permitted AI usage * permit AI for writing code
…gml-org#21270) * contrib : rewrite AGENTS.md, make it more clear about types of permitted AI usage * permit AI for writing code
…gml-org#21270) * contrib : rewrite AGENTS.md, make it more clear about types of permitted AI usage * permit AI for writing code
…gml-org#21270) * contrib : rewrite AGENTS.md, make it more clear about types of permitted AI usage * permit AI for writing code
7 participants
Overview
Alternative to #21007
This version make it more clear about the values of the project:
Requirements