Your requirements live in a chat window
Six rules from fifteen real OpenSpec changes for writing project docs an agent can't misread, including the one that says a proposal is allowed to fail.
Read the article
Nine years of Swift on apps people can’t really avoid using: health insurance, pharmacies, an identity card you verify over Bluetooth. These days a good part of that work runs through AI agents, and I write down what holds up and what quietly breaks.
Field notes on shipping iOS apps with AI agents
Claude Code against real repositories, hooks that hold, and the ten minutes of setup that decide whether any of it helps. Measured, not hyped.
More articles
All writingNotes
Now building
Agent Flight Recorder
A recorder for Claude Code sessions that runs on my own machine. It captures hooks, cost and timing, and the transcript, then rebuilds each session as something I can read: the agent tree, the context curve, where the turns went. Rules flag the patterns I keep paying for, like a loop of identical calls. It never sits between me and Claude Code; if it stops, nothing else does.
Projects
09
Claude Code — The Cheat Sheet (opens in a new tab)
Foldcard (opens in a new tab)
Through (opens in a new tab)
Prompts Lab (opens in a new tab)
Design Patterns (opens in a new tab)
Idea Clear (opens in a new tab)
QR Generator (opens in a new tab)
Quitify Now! (opens in a new tab)
Practice Words (opens in a new tab)