Ignore claude files: CLAUDE.md, .claude/

[yungyuc]

I am new to vibe coding. @tigercosmos @ThreeMonth03 @5000user5000 should I/we do this?

[tigercosmos]

We should not exclude these AI agent files. People should share and mentain the same ones.

[ThreeMonth03]

I'm not sure. I prefer vibe coding with cursor instead of claude code agent, so I haven't seen these files before.
By the way, AI often generates a lot of the example files or inrelevent files, and it should be removed by another prompt or by manual.

[5000user5000]

I'm a bit undecided, but I'd like to share my experience with this.

I primarily use claude.md as a form of "development memory" for the AI—tracking my progress through different phases and restoring context if the session disconnects. Since this content is dynamic and personal to my workflow, I agree it fits best in .gitignore.

However, I do believe we should commit a dedicated shared document for AI contexts (perhaps under a different name). This acts as a guardrail for new contributors, ensuring the AI is aware of our protocols—such as verifying CI and unit tests before pushing, or avoiding architectural pitfalls. For example, it could prevent mistakes like the one I made previously, where I implemented an overly long function directly inside a class declaration.

Suggestion: We should add claude.md to .gitignore for personal use, but establish a separate, committed file (e.g., docs/AI_GUIDELINES.md) to standardize these rules and reduce error rates.

[yungyuc]

We should not exclude these AI agent files. People should share and maintain the same ones.

@tigercosmos Makes sense. Would you please share what you use in a PR?

[yungyuc]

I'm not sure. I prefer vibe coding with cursor instead of claude code agent, so I haven't seen these files before.

@ThreeMonth03 Would you have time to share what you use with cursor in a PR?

[yungyuc]

I'm a bit undecided, but I'd like to share my experience with this.

I primarily use claude.md as a form of "development memory" for the AI—tracking my progress through different phases and restoring context if the session disconnects. Since this content is dynamic and personal to my workflow, I agree it fits best in .gitignore.

However, I do believe we should commit a dedicated shared document for AI contexts (perhaps under a different name). This acts as a guardrail for new contributors, ensuring the AI is aware of our protocols—such as verifying CI and unit tests before pushing, or avoiding architectural pitfalls. For example, it could prevent mistakes like the one I made previously, where I implemented an overly long function directly inside a class declaration.

Suggestion: We should add claude.md to .gitignore for personal use, but establish a separate, committed file (e.g., docs/AI_GUIDELINES.md) to standardize these rules and reduce error rates.

@5000user5000 This also makes sense, but contradicts to what @tigercosmos said. @tigercosmos, I'd like to know what you think?

[ThreeMonth03]

I'm not sure. I prefer vibe coding with cursor instead of claude code agent, so I haven't seen these files before.

@ThreeMonth03 Would you have time to share what you use with cursor in a PR?

No, I will fix the ongoing pull request and share the usage of cursor after the conference. Recently, I also take a lot of time to practice English Speaking with Toko, in order to attend the meeting by in English.

[ThreeMonth03]

I'm not sure. I prefer vibe coding with cursor instead of claude code agent, so I haven't seen these files before.

@ThreeMonth03 Would you have time to share what you use with cursor in a PR?

I have a suggestion. Maybe I could find a quite simple issue like pr #608 , then I fix the issue, record the process and upload to youtube. The length of the video is about half to two hours.

[tigercosmos]

Would you please share what you use in a PR?

I didn't use agent file for modmesh. Copilot works well with the default setup.

However I recommend to have a common AGENTS.md, which works for most IDES, and symbolic links to other agent files for different IDEs, e.g. CLAUDE.md.

Asking AI to read STYLE.rst or copy the content from STYLE.rst would be a good start.

[yungyuc]

Would you please share what you use in a PR?

I didn't use agent file for modmesh. Copilot works well with the default setup.

However I recommend to have a common AGENTS.md, which works for most IDES, and symbolic links to other agent files for different IDEs, e.g. CLAUDE.md.

Asking AI to read STYLE.rst or copy the content from STYLE.rst would be a good start.

@tigercosmos Having a comment set of files for hinting AI assistants will help align contributors. On the other hand, hard-coding to fixed file names would make it cumbersome when tryting new things, for it is easy to accidentally check in experimental changes to the hinting files.

Maybe the hinting files should go to a sub-directory, and contributors copy them manually to use.

@ThreeMonth03 Would you have time to share what you use with cursor in a PR?

I have a suggestion. Maybe I could find a quite simple issue like pr #608 , then I fix the issue, record the process and upload to youtube. The length of the video is about half to two hours.

@ThreeMonth03 sounds like a plan.