2026-02-09 21:02 UTC
@isaiah@mastodon.social I do this indirectly with the workflow I have set up for most of my tasks. I create a Github issue and basically put my prompt as the first comment in the issue, then tell Claude to read the issue and execute. For simple issues, I have it do its thing then commit and close. For more complex ones with a long plan, I'll tell it to save the plan as another comment in the issue to keep a kind of conversation history there, sometimes with multiple stages if it goes on a while.
Replies (1)
-
@isaiah@mastodon.social 2026-02-09 21:25
@bwebster@mastodon.social i think i'm not so much looking for a record of how i got here — but more of a way to move up an abstraction layer. but it seems like repeatability is needed — and repeatability doesn't see like a LLM priority. i guess for now it's still a manual interactive process on specific features. but i think the ability to treat prompts as the new "truth" and treat native source like assembly is not quite there yet. i suspect it's on it's way, though.