GitHub Repository Publishes Full Chinese Translation of Jev Engineering Notes for Coding Agents

A GitHub repository released a complete Chinese translation of "Jev Engineering: For Coding Agents," a work compiled from design notes by TypeSafe founder Diogo Almeida, preserving the original structure and seven illustrations. The text examines how coding agents should be designed without a KV cache and derives six design choices that existing agents inherit by default, covering model routing costs, context compression, sub-agent triggering, and token accounting. The repository also includes the author's original design notes for comparison with the finished document.
Event version 1.
Key insights
Gap lies in per-turn model input
The notes argue that coding agents largely reduce to a model plus a few tools in a loop, and that differences between products lie not in the loop but in what is fed to the model each turn.
Design premise without KV cache
The text starts from the question of how a coding agent should be designed if the language model has no KV cache, and derives six design choices that existing agents inherit by default.
Covered agent design topics
The material addresses why model routing fails to save money, why context compression is blind, and why sub-agents are rarely triggered, plus a section dedicated to token accounting.
Original notes included for comparison
The repository includes the author's original design notes, allowing readers to compare the draft with the finished document and see the processing from notes to final text.