TypeSafe ships official skill teaching Claude Code to batch Jev calls, cutting bills 12.2x
anselm · x · 2026-09-27
TypeSafe released an official skill that teaches Claude Code how to build on Jev correctly.
The problem: left alone, coding agents ask Jev one question per call and guess field names. The skill rewires the order before any code is written — docs → behaviour → judgments → one request → code decides — merging 13 questions into a single call, which TypeSafe's cookbook measures as a 12.2x bill reduction. The two-page skill breaks down the failure behind each of its 11 rules, offers a quick-pick table (Choice/Noul/Score), and a paste-ready prompt skeleton.
Failure mode warning: in a real incident, Jev sat between GPT and the author as a checker, but when it went silent, the agent decided silence was safer and stopped sending anything at all — a second agent had to unstick it. The author argues every checker needs a branch for silence: pass through tagged unverified, route risky spans to a bigger model, queue and retry, or wake a human for irreversible actions. A single Jev call takes 70-500ms; a half-second checker with no default still stalls the whole pipeline.
More from coding & agent
- FuSheng: AI now navigates human interfaces like a person and maintains real relationships — FuSheng_0306 · 2026-09-27
- AI agent sends personalized Mid-Autumn greetings to 50 contacts by seeing the screen and clicking, then replies to 70+ messages — FuSheng_0306 · 2026-09-27
- FuSheng had GPT-6 take over his WeChat and send 120+ error-free holiday greetings — FuSheng_0306 · 2026-09-27
- Pairing Astra and Opus for coding: 3-stage pipeline tops at 82.8, but solo runs cost far less — kevinkern · 2026-09-27
- UT Arlington paper: AI agents miss requirements even when marking tasks complete — alex_verem · 2026-09-27
- KeySync puts your 2FA codes in the Chrome toolbar, copying in 1.2s — aliscodes · 2026-09-27