AI coding rules
Agents go wrong at every stage: rules for AGENTS.md, model switches, long tasks, pages and launch.
By topic
- ReviewGet the work checked before you rely on it.
- CodebaseKeep a repo tidy while an agent works in it.
- DesignPages, interfaces and videos that look designed.
- Models and costThe model and effort that fit the job, for less quota and money.
- Long runsKeep an agent working for hours without losing the thread.
- LaunchGet a site out the door and in front of people.
- WritingGet prose that reads like a person wrote it.
Playbooks
Design taste for Opus 5.5
Pages look AI-made: a DESIGN.md from sites you like, components restyled, CSS checked.
Overnight long runs
Quota idles overnight: safe tasks to run, a HANDOFF.md and a progress dashboard.
Pre-launch checks
Launch day finds what nobody checked: pages, states, backend and user flows.
Quota and models
Quota runs out early: effort and models per job, the cache timed, API costs cut.
Repo rules for agents
Agents guess, overbuild and widen scope: rules for AGENTS.md and CLAUDE.md.