Why your AI agent forgets your repo every session — and the CLAUDE.md fix

Every new session, your AI agent re-discovers your project from scratch — reads the README, greps for the test runner, lists files. That round-trip costs tokens and time, every time. CLAUDE.md ends the loop. It's the one-page brief your agent reads at the start of every session.

What you'll learn in under 3 minutes:
→ Why a one-page file ends the Monday-morning cold-start tax
→ The four layers Claude Code loads at session start (project · narrow rules · personal · auto-memory)
→ Five sections that earn their place — Commands, Architecture, Hard rules, Workflow, Out of scope
→ The "#" shortcut that lets you write CLAUDE.md by living in your repo, not in a planning session
→ Three repos that extend it — claude-mem, everything-claude-code, awesome-claude-code

Source links:
→ LinkedIn (longer breakdown + 12-slide carousel): https://www.linkedin.com/in/-mazharulislam-/
→ Instagram (10-slide carousel + Reel, same series): @theainativeengineer
→ Official Anthropic docs on CLAUDE.md + memory: https://code.claude.com/docs/en/memory
→ Source post: Inside Claude Code series, post 04

#Shorts #ClaudeCode #AgenticCoding #AIEngineering #InsideClaudeCode

Видео Why your AI agent forgets your repo every session — and the CLAUDE.md fix канала theainativeengineer