anthropics
Built-in Learning Memory and Documentation Prompts
Built-in Learning Memory and Documentation Prompts
Priority: HIGH
Problem: Claude Code doesn't persist learnings between sessions or even within a session. Despite explicit user request: "please write details to Claude.MD so we don't need to go through this discovery again" - this was never done. All learnings were lost.
Current Behavior:
Discoveries made during debugging are forgotten No automatic documentation of solutions Same issues will be debugged again in future sessions CLAUDE.md files exist but aren't proactively updated Expected Behavior:
When CLAUDE.md exists, periodically offer: "I discovered X. Should I document this in CLAUDE.md?" After resolving a complex issue, automatically suggest documentation At end of session, summarize learnings and offer to save them Make documentation a first-class part of workflow Impact:
Wasted effort re-discovering same solutions No knowledge accumulation over time CLAUDE.md becomes stale despite being valuable resource Suggested Implementation:
Detect when CLAUDE.md exists in project After significant discoveries, prompt: "Should I update CLAUDE.md with what we learned about [topic]?" At end of complex debugging, offer session summary Make it easy to say "yes, document that" without interrupting flow Real Example: User explicitly requested documentation in transcript line 51, but AI never updated CLAUDE.md despite 700+ more lines of work. All iOS credential learnings were lost.
