The five-line handoff note: ending a Claude Code session so the next one starts fast
The expensive part of resuming a coding session is rarely the code. It is rebuilding the picture in your head: what the agent was doing, which approach you had already rejected, and what "done" was supposed to mean. With one session this costs a few minutes. With five sessions it can eat the first hour of the day.
A short handoff note, written before you stop, removes most of that cost.
The five lines
Ask the agent to write this at the end of every working block, in a file that lives with the work (for example NOTES.md in the worktree, or the task's issue):
- Goal — one sentence on what this session exists to finish.
- Done — what changed, with file paths or commit hashes.
- In progress — the exact point it stopped, including anything half-applied.
- Next — the very next concrete step, not a wish list.
- Verify — the command or check that proves the work is correct.
A prompt that works well: "Before we stop, write a five-line handoff: goal, done (with paths), in progress, next step, and the command that verifies it. Do not start new work."
Why this format
- It is short enough to actually happen. A long status report gets skipped on a tired evening. Five lines take the agent seconds.
- "In progress" is the line that saves you. Half-finished edits are where resumed sessions go wrong: the next session assumes a clean state and builds on a broken one.
- "Verify" keeps the next session honest. It gives the resumed agent a way to check the ground before adding to it, instead of trusting a summary.
- It survives tool changes. A plain file in the repository does not depend on any one tool's history format.
Resuming from the note
When you come back, start the session in the same directory and point it at the note first: "Read NOTES.md, run the verify command, and tell me whether the state matches before changing anything." If the check fails, you have found the problem before it spread.
Claude Code's own --resume restores the conversation, which is useful. The note adds the part a transcript does not give you quickly: a deliberate summary of intent. The two work together.
With several sessions at once
The note matters more as the number of sessions grows, because you are the one holding all of them together. Two habits help:
- Keep one note per session, in that session's directory. A shared notes file becomes a collision point — see when two AI agents edit the same file.
- Name sessions after their goal line. The name then tells you which note to open; see session naming conventions.
Getting the sessions themselves back after a restart is a separate problem. We built Agent Tile for that part: a one-time $10 macOS app that tiles each Claude Code session with its folder visible and relaunches each one with --resume in its original directory. It brings the sessions back; the handoff note tells each of them — and you — where to pick up.
Takeaway
End every working block with five lines: goal, done, in progress, next, verify. It costs seconds, it catches half-finished state before it causes damage, and it turns "what was I doing?" into a quick read.
Agent TilePractical guides for running many Claude Code sessions in parallel on macOS — orchestration, review workflows, and staying sane with a fleet of AI coding agents.
Get Agent Tile — full source, $10 once →
- IT-QA — Technical Q&A with AI-assisted answers and point bounties — for the questions that come up while you're running a fleet.