Tomas Toth

[Tomas Toth @kryptm4n]

@iamsahaj_xyz what's the font you are using?

[Tomas Toth @kryptm4n]

@joenjenga_ @_vmlops It got removed, check changelog

[Tomas Toth @kryptm4n]

@0xSero You should check https://t.co/QmiruvvQ8E (not full IDE, shows diffs though)

[Tomas Toth @kryptm4n]

@nicopreme bro do you even sleep :D

[Ido Salomon @idosal1]

My entire childhood has led me to this moment...

I built AgentCraft - orchestrate your agents with your favorite RTS interface! ⚔️

Coming soon 👀

[Tomas Toth @kryptm4n]

@dexhorthy I have tried multiple approaches for researching the codebase (including the generic one from your repo) and given it researches a backend feature that spans 10+ files it always misses some important details

[Tomas Toth @kryptm4n]

@claudeai

[Tomas Toth @kryptm4n]

@maxxyung11 why tf does someone need more than 1 laptop

[Tomas Toth @kryptm4n]

@trq212 More like, writing code with Claude: CRY

[Tomas Toth @kryptm4n]

@badlogicgames I mean you shouldn't blindly trust it with writing anything, to me, tests are not less important than production code. Maybe add a phase where CC comments out a line of code to see the test fail to verify it's working in the first place 🤷‍♂️

[dax @thdxr]

we figured it out

we figured out how to stop LLMs from inserting comments into code

but tell me, does the world deserve this gift?

[Tomas Toth @kryptm4n]

@badlogicgames I started getting it too today

[Tomas Toth @kryptm4n]

@badlogicgames And then I tell it to work on task by task (each task in separate subagent with clear context, where I tell it to pass the original plan md and the tasks json file) and tell it to proceed on next pending task

[Tomas Toth @kryptm4n]

@badlogicgames What I do is I have a comprehensive md plan containing what code changes should be done tests etc, and I tell CC to format that plan into a template file tasks-template.json (that I created originally using Claude Task Master, and it is working quite good)

[Tomas Toth @kryptm4n]

@steipete I like the idea of storing it in JSON so that you can tell CC to use jq for interacting with the file

[Siva Kovvuru @lazymindz]

don't start your next project without reading/watching all the nuggets from this article

https://t.co/j0W8UfLMRa

Thanks @steipete for the collection.

Learnt a few tricks for sure (logging centrally and tracing)

[Tomas Toth @kryptm4n]

@badlogicgames I wanna experiment more with TDD and ATDD, I think it might be a good way to give CC a feedback loop

[Tomas Toth @kryptm4n]

@badlogicgames Then I let it analyze complexity and create subtasks, after it's done I let it execute using a simple prompt using jq (thx for that), after each task it stops and waits for me to verify, after all is done it commits, I wipe the context (back to start) and repeat

[Tomas Toth @kryptm4n]

@badlogicgames Then I tell it to create a very detailed plan on how to implement the changes in the spec and I iterate with it to also include unit tests and BDDs. After few iterations when I am okay with everything I use taskmaster to create a plan it's a JSON file as well