Claude Code✓ Confirmed in official docsAdded Oct 15
How to have Claude Code audit CLAUDE.md, skills and memory written for older models
When you move to a new model, the CLAUDE.md, AGENTS.md, skills and memory you wrote for the previous one stay behind. They can make the model redo things it now handles on its own, or leave instructions that contradict each other. A post on X with 150 likes suggests having Claude audit everything it loads automatically and propose a cleanup, while keeping the pass read-only until you approve. Ask it like this.
Audit everything you load automatically or reference during tasks: CLAUDE.md, AGENTS.md, skills, memory and other instructions written for earlier models.
For each item, say whether it adds knowledge you lack, duplicates what you already do, or creates extra work. Flag conflicting instructions, obsolete tools, skill triggers that fire too broadly, and stale facts.
Separate confirmed defects from judgment calls. Treat file dates as clues, not proof. Keep useful domain knowledge and safety rules.
Recommend what to keep, simplify, verify or retire, citing files and examples, ordered by benefit.
Report here first. Stay read-only until I approve, and list anything you could not inspect.
In Claude Code, /memory shows which memory files are loaded and /context shows how much of the context window each part takes, so you can see what is heaviest. The same prompt works in Codex for checking AGENTS.md.
Get notified when a new tip lands
Tip alerts only go out in the app. Install it and turn notifications on.