@thsottiaux the fact that codex still can't ask questions outside of plan mode feels kind of crazy. It's not just useful for plan mode. I made a skill in claude used for step by step validation of changes I made that's perfect with the questions model. The best alternative in codex is annotating each validation check, but it's not nearly as seamless or organized as simply making use of a tool that already exists in the app.
@lofibanshee@RaiderIO_WoW Remember when the arcane tier set got nerfed by 5% and their discord immediately went into nonstop doom-posting and their head theorycrafter said all mage specs are just bad now? Mages are adorable man
@thsottiaux Allow agents to ask questions outside of plan mode. Make the 'undo' button actually functional. Make a built in 'caveman' mode where the agent speaks simply, concisely, plainly without special instructions.
1. this says literally nothing without actual profile information
2. plusmouse is outspokenly anti-AI and this is a thinly veiled attempt at throwing shade to EUI which is a largely AI generated addon
3. no one gives a shit if something is ai generated. they care if its bad.
4. trying to gaslight people that they're wrong for getting higher fps after switching to EUI is retarded
@thsottiaux Allow agents to ask questions outside of plan mode. Make the 'undo' button actually functional. Make a built in 'caveman' mode where the agent speaks simply, concisely, plainly without special instructions.
opus is doing great work for me so I don't fully understand the sentiment either. my guess is that people are still forcing skills/workflows/overwritten agent/claude.md's that were necessary for less intelligent models onto opus and finding bad results.
my experience with some of these skills/workflows is that they're way, way over-engineered. i think that's why the ceo of the group that made compound-engineering left a mid review; virtually all skills provided by that plugin are monolithic.
question for people who don't like opus: if you let the model cook more and hold its hand less, do you find better results?
my experience is a resounding yes
@thsottiaux allow agents to ask questions outside of plan mode. this is useful in instances where clarifying a direction is worthwhile without going through an entire planning phase.
@RohdeAli It was a lot of frontloaded work creating the skills and teaching the model to know the intricacies of the work, but now I just invoke a skill with raw data as input and it provides fully, professionally written reports. It's amazing.
@Hipposareneat Blizzard stores these glow textures as an atlas. What you're seeing is the entire atlas instead of each individual frame of the proc-glow animation instead of the correct glow at the correct moment. It's a baseline UI bug with assigning the correct textures
@thsottiaux The built-in code review is still a bit awkward to use. It was recently changed to be inline with the current session which is great since it has valuable context to base its review on but I still find that multiple independent sessions are needed to catch everything.