How to Share Context Between Claude Sessions

Share context between Claude sessions by saving a checked handoff with the goal, current decisions, completed work, source references, and one next action. Give that handoff to the new session and ask it to explain where it will start before it changes anything.

Claude’s recall features can help, but active work benefits from an explicit checkpoint that you can inspect. fenbs keeps that checkpoint with the project so the next connected session can read it without relying on a private conversation’s memory.

What should you save before the session ends?

1. State the goal in one or two sentences, including the result you are trying to produce.
2. List the decisions people approved and the rules that still apply. Separate unanswered questions from settled choices.
3. Describe what changed and where the current deliverable lives. Include file names or source references that the next session can actually reach.
4. Record what was checked, the result, and what remains unverified. Preserve failures and uncertainty as carefully as successes.
5. Name the next action and anything waiting on a person. Review the handoff yourself before treating it as the current work state.

For example, a nonprofit coordinator preparing a volunteer guide could save: “The welcome section is revised against the approved handbook. The contact list is still unconfirmed. Next, ask the coordinator to approve the two emergency contacts before updating that section.” That is more actionable than “We worked on the guide.”

Ask for the checkpoint while the details are still available, rather than after closing everything. A short review at the end is easier than reconstructing an entire afternoon later. If the assistant cannot identify a source or a check, keep that gap visible instead of encouraging it to fill the space with a confident guess.

Can memory and chat search handle this automatically?

Anthropic documents chat search and memory↗ (opens in a new tab) as ways to reuse earlier context. Availability and boundaries depend on account settings and the work surface, so check the vendor’s current guide rather than assuming every new session sees every previous exchange.

Recall is useful for locating an earlier discussion. It still needs a clear question: which project, which decision, and which version? “Find the wording we approved for the volunteer eligibility section” gives you something to verify. “Remember everything from yesterday” gives you no practical test of completeness.

Treat recalled material as a starting point for review. Check important statements against the current record, especially when you changed your mind during the old conversation. An early proposal can sound perfectly plausible even after a later message replaced it.

Our introduction to Claude explains the broader product. This handoff routine focuses on the work that must survive a session change, whether recall is available, disabled, or simply incomplete for the task at hand.

Where should the handoff live?

Choose a place the next session can access and you can maintain. For a one-time continuation, paste the checked handoff into the new chat. For repeated work, keep the current project instructions and reference material in a deliberate project home, with a separate checkpoint for changing progress.

Anthropic’s project management guide↗ (opens in a new tab) explains adding project knowledge and instructions. Use instructions for durable working requirements and reference files for the material they describe. Avoid putting a temporary “next step” into a rule you expect to use for months.

For a code repository, the record also needs the working state that exists outside the chat. Note the relevant files and verification results, then let the next session inspect the actual code. The Claude Code context guide covers that environment; passing a live or recorded session has separate considerations in sharing a Claude Code session.

When work moves among assistants, fenbs provides an outside home for tasks, Decisions and rules, and Where we left off. Keep long source documents in their appropriate storage and point to them from the handoff.

How much detail is enough?

Save what changes the next action. A complete account of every suggestion is rarely the most useful starting point. Keep the reason for a decision when it prevents the next assistant from reopening an option that was rejected for a real constraint.

For the volunteer guide, keep the approved audience, required sections, current draft, and open contact question. You do not need the ten alternative opening sentences unless one explains a requirement. A shorter checkpoint is useful when it removes noise without removing evidence.

How do you start the receiving session?

Begin with a read-and-confirm request: “Read the current project handoff. Tell me the approved decisions, what remains unverified, and the first action you propose. Do not change the deliverable yet.” Compare the response with your record, then correct any mismatch before giving permission to continue.

If a referenced file is unavailable, stop that part of the task and restore access or provide the approved source. Do not ask the assistant to pretend it has read the file. Missing context is manageable when it is visible; an invented substitute can quietly change the work.

Also check for work still happening elsewhere. A second session should not revise the same master document while another person is editing it without coordination. Name one current version and one owner of the next change, then save the resulting state when that turn ends.

How to do this with fenbs

1. Create a free account and add the project you expect to continue across sessions.
2. Add the current tasks and plans. Record approved Decisions and rules, and use AI context notes for facts the next session needs.
3. Connect Claude using Integrations. MCP is a standard connection for using outside tools; approve the intended access through browser sign-in.
4. Before stopping, ask the assistant to save progress for the tasks it touched and the project. Include completed work, checks, uncertainty, and the next action.
5. In the next session, ask the connected assistant to read the project context and confirm where it will begin. After it works, save the updated checkpoint again.

fenbs does not read your Claude history automatically or replace the files the work depends on. If you are continuing one short conversation once, a checked pasted summary may be enough. Use fenbs when the task record needs to survive repeated sessions or changes of assistant.

The AI handoffs topic and guide to MCP versus Claude skills explain related pieces of this workflow. Keep the first trial small: end one real task with an accurate checkpoint, then see whether the next session can identify its unfinished action.