Conversation
anandaroop
left a comment
There was a problem hiding this comment.
lgtm, just confirming one question below
|
|
||
| ## Code Style & Common Patterns | ||
|
|
||
| @docs/best_practices.md |
There was a problem hiding this comment.
The @ will immediately pull the linked doc (~4k tokens) into the context.
Just confirming that's what you want for best practices (as opposed to the further docs below, which will be progressively disclosed as needed)?
There was a problem hiding this comment.
Yes, that’s intentional. I did the math here and found the additional tokens would lead to 2% extra in the window size (assuming 200k size) but would make sure that the best practices are always respected. I initially wrote the best docs in a short format compared to the main best practices docs, but this would mean that we need to maintain parity. So for the time being, I guessed because it’s not a problem, yet, I will do the easy route
Description
In this PR, I am adding
AGENTS.mdandCLAUDE.mdfiles to make AI tools more efficient inside Eigen.How was this built?
1. Gathered best practices
Sources:
https://www.builder.io/blog/claude-md-guide
https://skills.sh/mounirdhahri/skills/agentsmd-claudemd-generator
https://acjay.com/2026/02/18/the-ai-landslide-is-happening/
https://agents.md/
Force agents.md https://github.com/artsy/force/blob/main/AGENTS.md
https://code.claude.com/docs/en/memory
2. Combined best practices into a skill
Using the above files and docs, I generated a skill to generate a
CLAUDE.mdandAGENTS.mdskill that we can use in our repos3. Manually review that it makes sense and that it follows our best practices and the community best practices. Mainly:
AGENTS.mdinsideCLAUDE.md4. Published the skill (maybe it's useful to someone else)
I published the skill here https://skills.sh/mounirdhahri/skills/agentsmd-claudemd-generator and installed it to generate the docs
5. Used the skill to generate
AGENTS.mdandCLAUDE.mdinside Eigen.6. Review generated docs and update the skill accordingly
What will be the impact of this on the context window?
~5,700 tokens on every conversation. That's still small — leaves ~194k tokens for actual work. For a file you'd want agents to reference on nearly every code change, that's a reasonable trade imo
If it ever feels too much, we can still reference
best_practices.mdfor example instead of importing it using@importPR Checklist
To the reviewers 👀
Changelog updates
Changelog updates
Cross-platform user-facing changes
iOS user-facing changes
Android user-facing changes
Dev changes
Need help with something? Have a look at our docs, or get in touch with us.