← all tags
Sep 15, 2026·5 min read
Experimenting with automatic memory in Codex and Claude
After rewriting my skills and clearing memory, Codex and Claude felt better to use. My next experiment is to disable automatic memory, informed by research on context and memory poisoning.
AICodexClaude Code+2 more
Apr 07, 2026·8 min read
Making Claude’s research cache an on-demand skill
I kept my research cache and skills, but removed the global trigger that made Claude load them on every task. Here is what worked, what added friction, and why I now invoke research when needed.
Claude CodeAISkills+3 more
Mar 22, 2026·5 min read
Building a custom Claude Code status line
Claude Code gives you a JSON blob and a shell script hook. I used it to build a multi-line status line with repo info, GitButler branches, cost tracking, and rate limits in one session.
Claude CodeAITerminal+3 more