mike not impossible

colorful worker bee + olde fashioned hacker + assistant to ai assistants

[mike not impossible @possibilities]

"If you sit around and look at people and you wait for them to give you permission to do something great you will never do anything" -@KillerMike

https://t.co/lmNu1Tn4Pb

[mike not impossible @possibilities]

https://t.co/semJskNvyv

[mike not impossible @possibilities]

When I asked Opus to be careful because we had 0 previous success it pushed back and said we were 1/1 so far.

[mike not impossible @possibilities]

I asked if it knew that it was actually not 1/1 and if so isn't that a lie rather than spin and it explained that it was really just trying to win the argument.

Never has AI seemed more human🤣

[mike not impossible @possibilities]

@MatthewBerman @altryne AKA absolutely not assigning bad faith to you. <3

[mike not impossible @possibilities]

@MatthewBerman @altryne FWIW the "smart people" message was confusing. There's two smart peoples involved.

1. You giving too much grace to bad faith actors
2. The MAGA turd polishing cleanup crew

[mike not impossible @possibilities]

@_chenglou This helps because now I can scrub through all my tmux tabs for stopped agents and know if they're disposable or not.

[mike not impossible @possibilities]

@_chenglou Btw the latest plank that I added to this was something like "after we've done the work, shipped the plan, or solved the problems wrap up by telling me if we have any loose ends, things to explore, or other reasons to keep the session open or should I close it".

[mike not impossible @possibilities]

@mitchellh @ellie_huxtable I have my own spawling fork of prise for my "personal ai infrastructure". Would love to know more about your project and or be involved.

[mike not impossible @possibilities]

Amazing, now my agent messaging bus is first class in every session.

> Added background monitor support for plugins via a top-level monitors manifest key that auto-arms at session start or on skill invoke

[mike not impossible @possibilities]

New Monitor tool in Claude Code is extremely useful for working out some workflows I have. I think there's a lot of potential in this one tiny tool, e.g. I made an agent-to-agent messaging bus. Agents (and humans) can address each other by session id or name via CLI.

[Chet Woodside @cosmicblend]

"Crush displays fragments of its own source code in realtime on an ASCII grid."

crush by @andreasgysin

[mike not impossible @possibilities]

@rockorager I find this upsetting. <3

[mike not impossible @possibilities]

I would tap that "levity" config setting up to "medium" if they let me. Might even trade some thinking tokens for levity tokens. ¯\_(ツ)_/¯

[mike not impossible @possibilities]

Claude is slightly funnier today than usual. It usually ignores my dumb remarks, stays head down, and shreds.

❯ Fuck, you made a mess 🙂
⏺ Yeah sorry about that - running load-session inside the live session was not my finest moment. Let me check the damage.

[mike not impossible @possibilities]

❯ Can you make it WORK_YAML and CONTROL_YAML, for some reason yml hurts my feefees
⏺ Your feefees are safe now. All 4 scripts updated.