ponytail makes your agent write one line.
puffer makes it write the framework that line was implying. YAGNI, correctly expanded: you absolutely gonna need it. +54% token spend. all gains, no chill. https://t.co/B46S1YwQlo @ClaudeDevs
Things I don't really like: reading the response, remembering everything I want to react to, scrolling to the bottom, then producing one neat answer.
Haiku autocomplete is handy and makes me less attentive, but autocomplete is trying to finish a thought. Often I'm still discovering the thought by writing the reply.
Maybe "chat" isn't it. A second of burst turns rich context into a one-liner. I'd like to think first, compress second.
I cope with Opus 5 using a duct-taped Neovim thing that pulls Claude's latest response with blockquotes when I hit ^G. Allows me to chime in directly under whatever needs a correction, "yes", "maybe", a question or a punt back.
I finish with a last touch: "STE-ifying" my scribble with a custom GPT into precise instructions. Opus kinda likes it, as prescribed: https://t.co/0XC3QvHafk
No scrolling back and forth.
@doodlestein acfs is now a single command: `/goal bv insights clean, marching orders sent, stop condition br ready --json=[]`
should i thank or blame you for the token burn?
Two things I deliberately did not do in this post:
1) properly thank @trq212, whatever is in the artifacts prompt, it works wonders
2) ask @claudeai to fix the punting in 5.1
say the word...
https://t.co/mpRZSdSo0C