If Claude Code loses the thread between sessions, you do not need a new tool. Five habits fix most of it. Everything below is generic; nothing to buy to use it.
1. Put what must survive in CLAUDE.md, and keep it short
Write facts the next session cannot guess (commands, folders, "never do X"). One line each, checkable. Delete lines that stopped being true. Check the current format in the official docs: https://code.claude.com/docs
2. Before a long job, estimate the cost
Write down how many steps, how many sub-agents, and what it would cost if it runs 3x longer than planned. If you cannot say, shrink the job. Use your plan's usage view if it has one.
3. Never accept "done" without proof
Ask: "Check it a different way than you did it (a second command, a test, a diff) and show me the output."
4. Stop loops before they burn your plan
Every sub-task gets a one-line goal, a limit (steps or minutes) and a stop rule ("stop if two tries give the same error"). There is a small free MIT-licensed skill for this: https://github.com/happy-program-zero/loop-stop
5. Hand over properly at the end of a session
Ask for 5 lines: what is done (with evidence), what is not, what is next, what is risky, which files changed. Save them as HANDOFF.md and start the next session by reading it.
This article was drafted with AI assistance.
Independent, not affiliated with Anthropic. "Claude" and "Claude Code" are trademarks of Anthropic.
Top comments (0)