01Links
Codex
Give the writing to the model I prefer for copy, while Claude checks every claim against its source.
Put your agent inside the repo, where it can read the plan, run the skills and check its work.
Why I keep it
I use Claude Code because it works inside the repo, where I keep the plan, skills, desks and checks. It reads the same files I do, so the instructions can stay with the work. Separate sessions each take a job in their own git worktree.
I keep the decisions; the agents do most of the typing. The tech-lead skill gives a build its routine, while /desk opens standing work from its saved records. Add the skills you need, so a new session starts with a method as well as a prompt.
01Links
Give the writing to the model I prefer for copy, while Claude checks every claim against its source.
02Links
Keep separate jobs open across Claude accounts, with Notch showing which account has room to work.
03Links
Speak the job, the look and the stack in one brief, so your agent gets the context together.
04Links
Carry a visitor’s voice-agent answers into a call booking, so they don’t have to start again.
I’ll send you one build log a week: what I added to the library and what I shipped with it. You can unsubscribe any time.