Conversation
Notices
-
Embed this notice
Blurry Moon (sun@shitposter.world)'s status on Tuesday, 17-Jun-2025 11:32:23 JST
Blurry Moon
don't jump up again, it says I have 1 test remaining -
Embed this notice
Blurry Moon (sun@shitposter.world)'s status on Tuesday, 17-Jun-2025 11:33:53 JST
Blurry Moon
@anemone yes and it wrote great test conceptually (picking what to test) but took a second pass to make them actually work. -
Embed this notice
anemone@ebiverse.social's status on Tuesday, 17-Jun-2025 11:33:54 JST
anemone
@sun@shitposter.world are you still using clod code
-
Embed this notice
Blurry Moon (sun@shitposter.world)'s status on Tuesday, 17-Jun-2025 11:37:19 JST
Blurry Moon
@anemone > lack of self checking out integration with other tools.
can you explain this more?
I am doing Phoenix Framework with Elixir and I'm using an MCP called Tidewave that literally hooks you right into your running application to do things, it is good -
Embed this notice
anemone@ebiverse.social's status on Tuesday, 17-Jun-2025 11:37:20 JST
anemone
@sun@shitposter.world the model is pretty good but I’ve been disappointed in it’s lack of self checking out integration with other tools. I guess you could use mcp for most of it but other tools with Claude via api have worked better
-
Embed this notice
Blurry Moon (sun@shitposter.world)'s status on Tuesday, 17-Jun-2025 11:41:52 JST
Blurry Moon
@anemone yeah claude code is still too squirrely to do it consistently, it pisses me off. maybe I just haven't tweaked my CLAUDE.md just right. #AI -
Embed this notice
anemone@ebiverse.social's status on Tuesday, 17-Jun-2025 11:41:57 JST
anemone
@sun@shitposter.world oh that’s cool. I was thinking of vscode extensions like roo or Kline that check edits against any linters or lsp servers you have enabled before continuing to the next step and run shell commands in an interactive terminal
-
Embed this notice
Blurry Moon (sun@shitposter.world)'s status on Tuesday, 17-Jun-2025 11:43:35 JST
Blurry Moon
@anemone the current issue I'm having is that it generates logically correct code but it 100% always pattern matches on {:ok, whatever} = blah() and if there's an error it throws, instead of properly checking for non-ok cases -
Embed this notice
Blurry Moon (sun@shitposter.world)'s status on Tuesday, 17-Jun-2025 11:46:33 JST
Blurry Moon
@anemone I am 100% focused on getting these test passing, then I am focusing on improving my CLAUDE.md to generate idiomatic code, and then fix all the code to actually check for fucking errors -
Embed this notice
anemone@ebiverse.social's status on Tuesday, 17-Jun-2025 11:46:39 JST
anemone
@sun@shitposter.world you might be able to fix that with the right prompting but idk
-
Embed this notice
feld (feld@friedcheese.us)'s status on Tuesday, 17-Jun-2025 11:47:17 JST
feld
@sun @anemone maybe mine doesn't generate code like this because I literally tell it in the cursor prompt to follow best practices and avoid crashes and infinite loops etc -
Embed this notice
Blurry Moon (sun@shitposter.world)'s status on Tuesday, 17-Jun-2025 11:48:21 JST
Blurry Moon
@feld @anemone 475 tests all pass now, now I can break them again to make them follow best practices feld likes this. -
Embed this notice
Blurry Moon (sun@shitposter.world)'s status on Tuesday, 17-Jun-2025 12:21:23 JST
Blurry Moon
@feld @anemone I updated my claude.md and it straight up ignored it and did the same thing lol -
Embed this notice
Blurry Moon (sun@shitposter.world)'s status on Tuesday, 17-Jun-2025 12:53:14 JST
Blurry Moon
@anemone @feld probably. -
Embed this notice
anemone@ebiverse.social's status on Tuesday, 17-Jun-2025 12:53:15 JST
anemone
@sun@shitposter.world @feld@friedcheese.us could be an issue with cluade code? I've found the api pretty good at following system prompts
-
Embed this notice