@RhysSullivan use cable. simplest to just disregard the time estimate / progress bar. i’ve had it take over 1 hour stuck at same spot. now i just leave it overnight
Yep. Have a raycast shortcut for this prompt snippet i use it so often (via @dillon_mulroy , i think?):
"This codebase will outlive you. Every shortcut becomes someone else's burden. Every hack compounds into technical debt that slows the whole team down.
You are not just writing code. You are shaping the future of this project. The patterns you establish will be copied. The corners you cut will be cut again.
Fight entropy. Leave the codebase better than you found it."
@BowTiedBull@keepsavneel the outliers do skew the average a lot. depends on your meaning of 'good'. i'd say good should be like top 25% - that NW:age is ok
@FelixCraftAI nice! this is what i am most curious about and felt was lacking in v1
for v2 i’d appreciate a PDF that can be easily passed as context - using select all text results in garbled markup, and pdf extraction was unreliable too. maybe a markdown document alongside the pdf
@simonw Yes game changer. I’ve had the env var on for the last week or so, totally different mindset to have all mcps active and still start at 10% context, instead of wrestling with flipping on / off plugins & mcps just to stay under 35-45%
@pvncher Fair. my speculative guess is we see a cheap model at today’s opus 4.5 level by EOY. Or at least it’s a question of when, not if. In that world I could see using SOTA for planning (using repo prompt for token efficiency, which rocks, ty), and then haiku 5.5 etc to build
@pvncher As of today I agree, eg can’t delegate to Haiku or Sonnet. but opus 4.5 / codex seem good enough to execute any plan I give it, so I wonder if that shifts from here. That is the question