Most AI coding skips design. Google shipped a plugin that refuses to.
EGM = design with an elephant, review with a goldfish.
Deep context for the draft. Fresh zero-context agents for the stress test. Then you approve before code.
I tested it. Solid. The independent reviewer pass is expensive on tokens. Worth it for real features. Overkill for a one-line fix. 😉
https://t.co/yfwfTTEXFD
New to AI coding? Skip the 50-tool lists.
Install this stack and start shipping:
Context7 — pulls current docs into the agent so it stops inventing APIs from 2023
OpenSpec — write the spec before code. Less thrash, fewer “rewrite everything” loops
TomViber — turns your agent into a tech-lead mentor: plan first, checkpoints, safe keys, and a PANIC button when things go sideways (EN/RU)
That’s it. Docs. Spec. Behavior.
Links:
https://t.co/B2oYYTnMAE
https://t.co/lSMPMjEziF…
https://t.co/y2prCSZO72
@thsottiaux One idea: let people buy a Pro pass for 3 days, a week, or two weeks. Sometimes you just need full power for one big project, but don’t want to pay $100 for a whole month. I bet a lot of people would pay $25 for a week of Pro instead of nothing at all. Seems like a win-win to me
Learning to build with AI? TomViber 🐱 teaches your agent (and you) to do it right from day one. Plans before code, checkpoints, safe keys, and a PANIC button. Free 👇
https://t.co/OATwcTzLY6
@thsottiaux Codex usage anomaly: security scan consumed nearly all weekly quota in ~25 minutes. Follow-up analysis with Luna consumed additional 4% in ~30 seconds. This appears disproportionate. Please investigate usage accounting for these runs.
@thsottiaux I really like how versatile it is and how all the essential features are right there. It’d be great to have some kind of built‑in memory for each individual project. And some example instruction sets for configuring a Codex agent.
@thsottiaux Tibo, since the latest Codex update, the limits have been disappearing way faster. Yesterday the same task took just 8% of my weekly limit, and today it jumped to over 30% with the same model. Also, Codex has been running much slower overall this week.