From Prompts to Plugins
A practical reuse ladder for Claude Code and Codex: turn repeated prompts into always-on instructions, on-demand skills, and installable plugins.
A practical reuse ladder for Claude Code and Codex: turn repeated prompts into always-on instructions, on-demand skills, and installable plugins.

A practical guide to Claude Code agent loops: how /goal, /loop, hooks, subagents, agent teams, memory, and dynamic workflows fit together.

I am tinkering with a trading-card-style project about debt, starting with university stats, shiny finishes, and one extremely loaded comparison.
Agents make a new top-down companion workflow possible: start with a detailed plan, turn it into rough working software, then shape the result by judgment, pruning, and proof.
Having trouble setting up Codex remote control? If the localhost callback fails with HTTP 431, oversized cookies from unrelated local apps may be the cause.
Hosted macOS Actions, a self-hosted Mac runner, TestFlight rehearsal, and physical iPhone receipts for honest Apple CI.
Machine Learning Refined resurfaced on social this week, a nice reminder that from scratch is not just a teaching style. It is how the book, code, and exercises were built.

Wispr Flow is a nice wrapper around speech-to-text models. That is a real product. But wrappers live or die on trust.

A meme about AI coding agent verification: treat “done” as a claim, not proof.

A real iteration log: designing a macOS app icon with Claude Code and the logo-designer-skill. 20 SVG iterations, 6 parallel subagents, and the surprisingly hard problem of curving a sound wave around a cylinder.