draht /dʁaːt/ Longform on AI coding agents, freelance engineering, TDD, DDD, and the pieces we solder together along the way.
A capable model still fails in characteristic ways. These eight disciplines — read the real request, decompose into checkable pieces, spend effort where the risk is, re-derive claims, label known vs guessed, attack your own conclusion, answer first, know the mistakes that look like competence — are now woven into every Draht agent, command, and skill.
Draht's May release adds a STATUS protocol on every subagent, a spec-reviewer that gates code-quality review, a four-phase /fix debugging protocol, and Red Flag stops across the workflow.
How Draht's GSD workflow commands use atomic reasoning to decompose complex problems into verifiable units before execution.
More articles coming. Follow along on GitHub.