@theo@cdxker@t3codes@bot Modern models are good at answering side questions in-thread and then continuing without interruption. But when I have several threads active, I often want to ask a question, go look elsewhere while it's being answered, and come back to it later.
@mattpocockuk "Content mappers" is a funny name for "full metaprogramming and executable macros". Ridiculously cool what this enables, such as custom DSLs and languages that compile to TypeScript with built-in source mapping.
@thorstenball "model answers and then continues" and then the answer gets lost in the transcript. I have multiple agents going all the time so I will often ask a question and come back to look at the answer later
@zeeg You're right to point this out. You shouldn't have to keep reminding me to process your emails as they come in. If you'd like, feel free to step away and come back to this later. The emails will still be there when you return.
@bytesalad@RhysSullivan It's compositional. You can just define a schedule once and then reuse it once with `anyEffect.pipe(Effect.retry(schedule))`
https://t.co/LzLTitMIJm
@TheRohanVarma Make them work properly with auto review. Threads my dot launches keep getting blocked by auto review because the message from the dot looks like a tool call, not a direct user authorization.
@TheRohanVarma Whenever my dot tries to interact with a codex session auto-review denies it because of "an authorization-provenance problem: the reviewer sees the coordinator’s relay, but doesn’t accept it as evidence of your original instruction". How are you supposed to use this?
@TheRohanVarma@a4ihd7vOmCAv0cs My dot literally can't spin up a local thread. It's connected to my desktop but it seems to only be able to start and interact directly with cloud threads. What am I missing?