The economics of LLM caching seem entirely broken. I often run a Monitor(heartbeat 4.5min "cache keepalive") with claude code to avoid being hit with massive quota costs every time something takes more than 5 minutes. The crossover period (for cost) is like 40 minutes, so it's kinda on-par with 1hr caching but costs less overall (assuming tasks take like 10 minutes or whatever). This is also why bg subagents and tasks are the way to go. This will only get more useful if 30 min caching becomes the norm, where the 2x cost crossover will be 4hrs, so it's like basically free to keep all your agents cached all the time.

friends don't let friends `echo "speak like you smoked toooooooooo much weed." >> ~/CLAUDE.md`

anything․sh - an #autopoietic #bash script I put together that grows itself as you use it. From `$ anything․sh "install docker"` to `$ anything․sh "skyrim action adventure RPG, pretty interface, thematic UI"`, it does anything! xertrov.github.io/anything.s…