Eric Provencher (@pvncher) just dropped a really solid write-up on rethinking skills, AGENTS.md, and prompts now that GPT-6 Astra is out. Short version: a lot of the scaffolding we piled on over the last year is now working against us. Bloated skill descriptions, "read the whole repo first" habits, over-specified recipes, and boundaries that make Astra stop too early. Worth a house-cleaning pass. Go read it: https://x.com/pvncher/status/2095991462416490862

Going live at 12:15. GPT-6 Astra just dropped and almost nobody has real hands-on yet. I’m stress-testing it the only way that counts, building real apps live, in depth, until we hit a wall. Benchmarks look wild on paper; I want to know what happens after hour two when the context gets messy and it has to hold a whole codebase. Hang in chat, throw ideas, and I’ll call it straight where Astra beats what I’m already paying for and where it doesn’t. Watch live: https://youtube.com/live/kc-azZ5z4c8

Lauren Tan just published part 1 of a written guide to pstack. This is not the Maven recording. This is how she actually uses it. Part 1 is verification. The agent has to check its own work so you are not the bottleneck. She treats that skill like infrastructure. Install pstack, run /create-verification-skill, and give the agent a small CLI it can drive, not just markdown. She has been doing this as the gardener on Grok Bot while the team lands hundreds of PRs a day. Volume only matters if the quality holds. Read it: https://x.com/i/article/2094151284949688320

Alex Lavaee and I did two masterclasses on this. This is the edited cut under 40 minutes so you can finish it. Your coding agent can produce a diff. You still decide what runs, what counts as done, and what evidence is good enough to ship. Alex walks through control loops and graphs, Atomic as a verifiable coding agent runtime, context as a budget, artifacts instead of dumping transcripts, durability and mid-run steering, and turning skills into inspectable workflows. Proof beats the model saying it is done. Watch: https://youtu.be/epg292lGCZ4 Atomic: https://github.com/bastani-inc/atomic

I threw away the whole app and rebuilt it from scratch. I didn't lose much because I wrote the spec down first. The Mac app was working and connected to the cloud. Then I realized I don't want to charge a subscription for something the new local models can run on device. Cursor read everything and made a spec. That spec is the recipe. Bitrig drives Claude Code and Codex on SwiftUI rails, builds in phases, and verifies each phase. Live mic and demo are working. Members can watch now. Live for everyone at 7am PT. Watch: https://youtu.be/wsSkjEmPVHU