@kunchenguid There is still one tradeoff with higher reasoning: Context grows much quicker. This might be a problem depending on the task and the model. I still like to keep this in my mind when choosing the level. Otherwise great insights and advice
@dillon_mulroy@thdxr I think subagents are less token efficient in general - says my gut.l - but they help you to have one orchestrator achieving a goal and not losing its mind after compactions
Kudos @doodlestein for DCG. Adding a custom rule for `git worktree prune` was so simple I didn’t even need to bother you with an issue. Keep up the great work. Thanks a lot!
Context: my sandboxed agent messed up my worktrees badly 😔
@doodlestein@amorriscode It always depends on your use case. If you are working with others and not pushing to main directly, it is a good tool to use. In a more solo dev use case they are overkill. Experienced this myself
what a great design choice. if only I gave Claude a skill which it could use to remember stuff. Oh wait I did, it just ignores it if not explicitly prompted. Codex models like to use the skill
confirm that your memories are broken with worktrees
Claude: Yes, confirmed. The memory system is broken for worktrees. Each worktree path gets its own isolated memory directory, so nothing carries over between them. It's a fundamental limitation of the path-based storage design
@rudi_anschober@WKogler@sebastiankurz Wieso können die Ausgangsbeschränkungen nicht ab sofort gelten? So liegt ein Wochenende vor uns, an dem sich unvernünftige Leute erst recht noch einmal sehen. Das kostet wieder Kraft,Energie,mehr Zeit im Lockdown und vor allem Menschenleben
We're giving away a Fathom cat hoodie. They are in super high demand and cannot be bought anywhere. And this will be the only time we do a retweet competition for a cat hoodie.
RT to win. Good luck 😃
@kettanaito Try ctrl-shift-t (cmd for osx) normally works for me even after restarting the computer. It opens closed tabs, but can work on whole windows.