Add AI agent guidelines (AGENTS.md)#146
Conversation
| @@ -0,0 +1 @@ | |||
| AGENTS.md No newline at end of file | |||
There was a problem hiding this comment.
Should this be @AGENTS.md?
ref: https://code.claude.com/docs/en/common-workflows#reference-files-and-directories
There was a problem hiding this comment.
In this setup, CLAUDE.md is a symlink to AGENTS.md. This makes the configuration agent-agnostic, allowing it to work with different popular LLM providers. This is done cause Claude Code doesn't respect AGENTS.md.
There was a problem hiding this comment.
I was today years old when I learned that github displays symlinks as regular files with the target path as the file content.
The convention I've seen is to use reference directives in the markdown.
For Claude, we would just create CLAUDE.md as a regular file with the following content:
@AGENTS.md
There was a problem hiding this comment.
Yes, this will work too.
| 2. **Walk through the code**: Explain step-by-step what happens | ||
| 3. **Highlight a gotcha**: What's a common mistake or misconception? | ||
|
|
||
| Keep explanations conversational. No newline at end of file |
There was a problem hiding this comment.
This file is exclusive to Claude. Is this what we want?
There was a problem hiding this comment.
This is the very first SKILL, created just for the demo. Claude was chosen as one of the most popular vendors.
There was a problem hiding this comment.
Seems like we should come up with a more agnostic scheme for defining these types of things (unless we're 100% in on Claude).
There was a problem hiding this comment.
Mind you, this is a draft PR. This work was not requested. This can serve as the foundation of what we build on top of that. This PR is not meant to be approved/merged immediately. There are more considerations to take care of before we move forward, but hey kudos @pavliuko for starting the fire
There was a problem hiding this comment.
in fact we also need a language-agnostic scheme as a first layer, enhanced in a second layer with a language-bound scheme
There was a problem hiding this comment.
This is a great question to discuss. Do we want to be vendor-agnostic or not?
This approach has a major disadvantage. It requires a lot of time and effort to support compatibility with different vendors. From my experience, instruments evolve very quickly and continuously and it makes difficult to keep up your setup updated.
On the other hand, it gives the opportunity to adopt new instruments. Vendors keep adding and updating them all the time. It also allows us to switch between llms to compare them and choose the best one.
3 participants
Establish a single source of truth for AI-assisted development guidelines.
AGENTS.mdserves as the shared config for all AI coding tools (Claude Code, Cursor, Copilot), withCLAUDE.mdsymlinked to it.Summary
AGENTS.mdwith project overview, agent protocol, git rules, build/test commands, architecture, and code style guidelinesCLAUDE.md→AGENTS.mdso Claude Code picks it up automaticallydeveloping-code-explainskill