How to Stop Claude in Chat and Resume the Work Later

Claude has a Stop button. While a response is generating, you will see a stop or square icon in the input area. Click it and the generation ends. That is the direct answer to stopping Claude in chat — and there are two useful things to do next: redirect the request, or save your progress and close the session.

Stopping a generation does not erase the work. The partial response stays visible. You can quote it, refer to it, or ask Claude to continue from a specific point. What it does lose is the current session context when you start a new conversation. Tools like fenbs keep the work checkpoint outside any single chat, so resuming later is a question of reading a record, not reconstructing a memory.

Stopping a response that is heading the wrong way

If Claude has started generating something clearly off-target — wrong format, wrong depth, misunderstood the task — stop it early. Letting it finish wastes time and can make the next correction harder because Claude tries to continue the direction it took.

After stopping, send a corrected prompt. Be specific about what was wrong: "That was too long. Give me a three-sentence version instead." or "You misunderstood — I want the analysis from the buyer's perspective, not the seller's." The more specific the correction, the less likely it is to drift in the same direction again.

In Claude Code, you can interrupt a running task with Ctrl+C in the terminal. Claude Code will stop the current action and wait for your next instruction. This is useful when a code generation or file operation is taking longer than expected or going in the wrong direction.

Stopping to save progress before closing a session

The most useful thing to do before closing a Claude conversation is to ask Claude to summarize where things stand. Ask: "Summarize what was completed in this session, what remains, and what the next step is." Copy that summary somewhere outside the chat — a document, a note, or a project tool.

Without that summary, the next session starts from zero. With it, you can paste the summary at the start of the next conversation and Claude picks up in seconds rather than minutes.

If the conversation has grown long, stop before you reach a natural break and ask for a compact summary: the goal, the decisions made, and the next step. A single focused paragraph is more useful to a future session than a page of detail.

Interrupting a very long task mid-way

For longer tasks — a multi-section document, a code refactor across several files, a research outline — it is better to structure the work in stages than to ask for everything at once. Ask for the outline first, review it, then ask for each section in turn. Stopping and redirecting is easier at the outline stage than after several thousand words.

When you must interrupt a long task partway through, stop the generation and ask Claude what it has covered and what remains. "You stopped generating mid-task. What did you complete before I interrupted, and what is left?" This re-establishes shared understanding without reading back through everything.

When to start a new conversation vs continue

Not every stopped session needs a new conversation. If you interrupted to redirect Claude or to add more context, stay in the same conversation and continue. The history is still there, and Claude can use it.

Start a new conversation when the current one has grown so long that Claude is losing track of earlier instructions, when a task has naturally ended and a different one is beginning, or when you want to approach the same problem with a completely fresh framing.

When you do start fresh, the first message matters. Paste in the decisions made, the current goal and any constraints that apply. A well-written opening message to a new conversation is faster to produce than it sounds: you are writing what you already know, not discovering it again. If you use fenbs, ask Claude to read the project context instead — that opening message writes itself.

How to do this with fenbs

1. Create a free account at fenbs.ai and add your project with the goal and current tasks.
2. Connect Claude at /integrations using the guide for your client.
3. Before stopping a session, ask Claude to save progress: mark completed tasks as done, update Where we left off, and note the next step.
4. When you resume, open a new conversation and ask Claude to read your fenbs project context. It will read your rules, current tasks and Where we left off before starting any work.
5. For stopping a generation that is going wrong: use the Stop button, send a corrected prompt, and continue without losing the project context that was already in the session.

For stopping a Claude Code task in the terminal, see how to stop a Claude Code task. For the related question of what happens to chat history when you close a session, see does Claude delete chats.