Juli

building

[Juli @0x_juli]

@rezoundous https://t.co/YHlsMIY5qv

[Tibo @thsottiaux]

To calibrate you all on which reasoning effort to use for Astra, know that GPT-6 Astra on low performs better than GPT-5.6 Sol on high.

If you were using high reasoning efforts with Sol and were happy, I suggest you move down to low or medium for Astra.

[Juli @0x_juli]

@rezoundous low is a better default for most people

[Juli @0x_juli]

Instructions that have helped my agent (put this in your system prompt or global AGENTS.md):

“Lead with the answer and include the context needed to act or understand”

“Sentence fragments are encouraged when they save words without reducing clarity”

“Prefer the shortest answer that preserves meaning, needed context, and material caveats”

“Apply Orwell's rules throughout: use short, familiar words, cut needless words, prefer active voice, and avoid jargon when plain words work”

“Let text-based visuals carry the explanation. Add only the prose needed to support them, or use prose alone when clearer.”

Visuals have had the greatest impact on clarity. It helps to specify useful forms: trees for structure, diagrams for data flow, and flowcharts or pseudocode for execution. When comparing designs, mark additions with +, changes with ~, and removals with -.

[Juli @0x_juli]

@owengretzinger 1) tmux
2) ghostty
3) nvim + git-delta
4) systemd/launchd

[Juli @0x_juli]

@stolinski This plugin is great, but I’ve found you need to edit the config to avoid some of the cache invalidation drawbacks.

https://t.co/oH5wlEy22E