yeah. self-review is theater. we ran agents copying seeded shortcuts from nearby files — 1,125 local trials / 15 fixtures. docs still let ~43% through; pre-write eslint: zero. published cut was 140/7. working on larger multi-model runs now. also building checkyourvibe — comes opinionated with rules from the get go, pre/post gate hooks, rules easy for agents to add for whatever you call good code. gate stops the bad write before it lands and tells the agent not to take those shortcuts. Agent-Native Evasion Defense: rules that cover each other so a cheap evasion trips the next rule
https://t.co/U9nxq4KqNQ
https://t.co/V45pdJEzQO
@mikepat711 unified vendor sub would be nice. until then cyv keeps concurrent work on the subscription clis you already have (cursor/grok/etc) and routes across them. cli, open source feedback/PRs welcome
https://t.co/RHLICxnjLX
that's the pain cyv is aimed at. multi sub / multi provider lanes on flat cli seats instead of juggling accounts by hand. still not one vendor mega-sub — but it discovers the clis you already pay for and spreads work. if that $1k problem is real for you, try it and tell me what's missing
@adrianhqs built checkyourvibe for that. grok/claude/codex/agy/devin — one harness, pre-edit gates so they write how you want, load balanced across whatever clis it finds. open source, looking for people who'll break it / PR it
https://t.co/RHLICxnjLX
I made checkyourvibe for this reason. Grok,Claude,codex,agy,devin. All harnessed and forced deterministically write code the way you want. Nobody seems interested and I'm running devins swe almost 24/7. Agy gets used until it runs out, pretty quick but more usage than the 20$ I pay for it. Same for codex, Claude, and grok. Cyv balances the load across whatever clis it automatically discovers.. and much more actually. It's a cli. Open source and I'm looking for help from people who get it
not assuming. measuring. if you run agents hard and want something that denies the bad write before it lands + keeps you off metered api — come break it. PRs welcome.
also: concurrent agents on api tokens will eat you alive. cyv keeps work on flat subscription cli lanes (multi provider/model) so the bill stays a seat not a meter. looking for people who'll actually use it and yell at me when it's wrong.
ran 140 trials on agents copying shortcuts from nearby files. docs helped a bit. eslint gate before the write almost wiped it. same idea cyv is built on — if you're fighting that too, try it / open a PR
https://t.co/RHLICxnjLX
Is it just me? It seems grok cli, cursor cloud agents and bots all eat the same usage meter? So what's the grok heavy limit I've not touched? What's up @bot?
@bot I'm offloading 90% of my work to other agent CLIs but I'm still burning through 20% of my bot budget a day with what I want to call pretty light usage. Probably on me but that still doesn't really work.
@TTrimoreau Napster, and grooving to whatever music you wanted for the very first time and that feel you just can't get without swat teams raiding your parents house.