I ran this prompt against an internal tool and it found and then fixed over 50 issues. A lot of small little things but also some huge workflow flaws.
Highly recommend you give it a shot!
Vibecoders, 𝗜'𝗺 𝗯𝗲𝗴𝗴𝗶𝗻𝗴 𝘆𝗼𝘂, open your messiest project and paste this into Claude Code or Codex:
"Find all of the implicit state machines in my app. Make the best decision that will elevate all three: the user experience (UX), developer experience (DX), and agent experience (AX) always, without breaking anything."
Claude Code / Codex tip: one ordinary email signup form, drawn as a state machine, has 𝟰𝟳 𝘀𝘁𝗮𝘁𝗲𝘀 𝗮𝗻𝗱 𝟭𝟮 𝗯𝘂𝗴𝘀.
Have it draw the state machine hiding in your code too. Paste this 👇
"Treat the repo as read-only. Pick the feature with the most states and draw every state and transition it has, from the UI to the server, as a single HTML page that fits on one screen. Mark the transitions that shouldn't happen in red and number them. Underneath, list the file, the line and why each one is wrong. For any you can reproduce, write a small script and run it."
@jeremybernier Claude Code is just the latest phase. Next year the timeline will be “I only talk to one agent and it yells at the other 19 for me.” Friday energy.
@poseidanai I don't normally have an issue with my agent forgetting things but it does happen from time to time.
Summarizing and unfolding only when needed is a practical way to stretch a session. Have you noticed any quality drop on the details that matter?
@zyonbuilds This is the kind of small skill that actually gets used. Describing the stack in plain English and getting a diagram is a nice shortcut for READMEs.
@TheViableEdge I have not messed wiht it yet but voice improving is the part I’m watching. Once that feels natural, a lot of the other friction gets easier to live with. I find voice ability very useful.
How fast would you learn about a critical CVE on your own PC?
Mine takes one night. I built a scanner into my homelab, and 26 Chrome criticals just proved it works. The full breakdown is below. https://t.co/ntTZNJA83v
@itsarbit Right now in testing I'm reviewing all changes manually. There seem to be WAY more than I expected. Adding an allow list per vendor is actually a great idea. May have to mess wit that.
Background subagents in Claude Code now use less of your usage limit.
A running subagent updates its status line every 30 seconds, and each update used to re-read its whole conversation. Now Haiku 5.5 writes it from short notes instead, so each update costs about 99% less.
@itsarbit Only tool calls honestly. Anything written into ~/.claude between sessions would slip past it. Closing that gap means hashing the dir at session start and diffing against the last known good. Good catch, that's going on my list.