Davidiusdadi

Thanks for creating this.

Would be interesting to know how much a jj-specific SKILL.md would race the score.

Maybe this does not fit the evaluation framework but I'd still be interested in your experience / setup with e.g. a terminal based coding agent such as claude code.

thanks for sharing

I personally feel most graph patterns are just easier implemented using functions unless the graph has some other purpose like observability or even just visualization also those can always be added ;)

I prototyped a similar system at a hackerthon last year: https://github.com/Davidiusdadi/obsidian-node-canvas

Your docs would be more readable if you made the code blocks just wide enough so that the lines don't wrap.

https://pocketflow.ai/ whitepaper link is 404ing btw.

Since they support so many databases the integration with specific providers is very shallow - e.g. they do not allow to make use of SQL views.

I personally found the typescript support disappointing in general but specifically around strict null handling. Their graphql schema was also declaring far to many fields as nullable which was a deal breaker for me looking for at least one way to generate precise typescript typings.

The same idea has implemented a while ago by http://readlang.com/

Being able to repeat the words i "looked up" is essential for me.