[bredren]

If you decide to add a utilities section, please consider FileKitty.

https://github.com/banagale/FileKitty

Despite all the hoopla of “project knowledge” and supposed codebase-wide context, I still find reasoning models do their best when directly provided with files relevant to a problem and nothing more.

I plan to add a tree feature and restore some other features I had in prior versions.

There are probably other tools that don’t require completion API requested but assist in AI enhanced dev workflows.

[dimitri-vs]

I actually hate using the AIs on third party platforms/wrappers because you have no idea what garbage they are putting in the system prompt and they have a strong incentive to minimize costs on LLM tokens. The difference with the quality I get just dumping all relevant context into the biggest SOTA model with reasoning tokens maxed out is massive.

Its to the point where if Cursor gets stuck on something I will just tell it to give me a recap of the details of the issue and suggest relevant files and then paste that into the Anthropic Console with thinking tokens set to max.