Nov Claude Code Tips

Day to day, set the model intensity only to Medium rather than chasing the highest level, and hand most development tasks to Claude through conversation.

Long tasks tend to get lazier the longer they run. You can use a script to remind it every 30 minutes, and state in the rules that if it makes no progress over two rounds at a stuck point, it should take over.

Have the Agent map out the impact scope and draft a plan first, then run tests after writing, and review the Diff with a clean context to reduce self-confirmed errors.

Use the constitution, specify, clarify, and plan commands to clarify the specs and open questions first, then let the AI start writing code.

Write colors, spacing, components, and design rules as layered documents, so the Agent doesn't have to guess when building interfaces, and can better match personal taste.

A single npx skills add command installs a chart or aesthetic Skill, letting the Agent directly produce flowcharts and more polished pages.

When the AI can't see geometric results, translate errors into compiler errors and check numbers, and it can find and fix problems on its own.

When you see a prompt saying permissions are disabled, don't assume the account is banned at first. Check in account settings whether the subscription has expired without renewal and reverted to the free plan.