Compact a Claude Conversation Without Losing the Work

To compact a Claude conversation safely, prepare a concise record of the goal, approved decisions, current sources, and unfinished work, then check it against the originals. In Claude Code, /compact is a documented command for summarizing the session. In a browser chat, you can ask for a continuation note and review it before starting fresh.

1. Pause at a meaningful checkpoint and identify the next task.
2. Ask for a continuation summary that separates approved choices from proposals.
3. Compare the summary with the actual source material and results.
4. Save the reviewed decisions and progress outside the conversation.
5. Resume with the summary and the sources needed for the next step.

Tools like fenbs keep that reviewed project record outside Claude, giving the next connected assistant a stable starting point.

What does compacting a conversation accomplish?

Compacting replaces a longer discussion with a smaller representation of what matters. The result should help an assistant continue the task without replaying every draft, false start, and repeated instruction. It is a selection of information, so checking what was left out is part of the job.

Think of a designer reviewing a product brochure. The next session needs the approved audience and current draft, plus the fact that the legal wording still awaits review. It probably does not need every unused headline. But if the summary drops the pending review, making it shorter has changed the project’s meaning.

Keep the source conversation or original documents available while checking the note. A summary makes an entry point; it does not become stronger evidence than the material it describes. For background on the assistant itself, see what Claude is.

Is the compact command available in every Claude app?

Anthropic’s Claude Code cheatsheet↗ (opens in a new tab) documents /compact and optional focus instructions. Keep that guidance attached to Claude Code, the coding product, rather than assuming a command has the same behavior in every Claude interface. The guide to Claude Code explains that product’s role.

For ordinary Claude chats, Anthropic’s usage and length guidance↗ (opens in a new tab) describes automatic context management when code execution, the setting that lets Claude run code, is enabled. Earlier messages can be summarized as the conversation grows. Availability and conditions belong to that documentation; you do not need to assume a universal browser compact button.

A manual continuation note is still useful as a review step. You can ask for the note in plain language, check it, and use it when moving to another conversation. Automatic context management does not replace the need to identify who approved a decision or whether a required check actually passed.

What should you tell Claude to preserve?

Give the summary a structure instead of asking only to “summarize everything.” A general recap can spend space describing the discussion while omitting the exact next action. The request below keeps attention on the state of the work and asks the assistant to expose missing evidence.

Prepare a continuation note for this task. Include:
1. Goal and definition of done.
2. Decisions a person approved, with reasons and references.
3. Continuing rules, preserving exact wording when necessary.
4. Current files or documents and their versions.
5. Work completed and checks actually run.
6. Unfinished work, uncertainties, and pending approvals.
7. The first next step.
Separate suggestions from decisions. Do not infer missing results.
List any important detail you cannot verify from the available context.

In Claude Code, you can use relevant focus instructions with the documented command. For example, ask the compaction to preserve the files being changed, the agreed scope, and checks still required. Treat those instructions as a way to guide selection, then verify the continuation instead of expecting perfect preservation.

In fenbs’s handoff workflow, Where we left off preserves the current task state alongside the decisions that should govern it. That lets you keep changing progress without repeatedly rewriting the person’s original choice.

How do you check that the summary kept the meaning?

Review the claims that would change the next action first. Compare approvals with the person’s actual words, completed work with the current deliverable, and test results with the recorded check. A fluent summary can still turn “proposed” into “agreed” or “drafted” into “finished.”

Keep precise identifiers and important phrases intact. Shortening a filename, source title, or acceptance requirement can make a note ambiguous. Remove repeated explanations around those details before abbreviating the details themselves.

What should stay outside the compact summary?

Keep long reference documents, full logs, and original deliverables in their appropriate homes. The summary needs enough information to locate and interpret them, with the relevant excerpt when the next action requires it. A list of inaccessible references is not a usable handoff.

Avoid treating the summary as a place to make new decisions. If two earlier messages conflict, flag the conflict for a person instead of silently picking one. If the current source is unclear, identify the versions and leave the choice open until someone resolves it.

Keep secrets and sensitive material out of reusable notes when they are unnecessary for the task. An instruction can name the approved source without repeating its confidential contents. fenbs’s Decisions and rules are useful for the agreed work boundaries, while the source material stays where its access is managed.

How do you resume without repeating the whole discussion?

Start by giving the next session one bounded action. Supply the checked continuation and the material it needs, then ask the assistant to restate the current goal, governing choices, and remaining uncertainty. Fix misunderstandings before asking it to change the deliverable.

Try a small continuation first, such as revising one section or inspecting one result. Check that it respects the preserved requirement. This is a more useful test than asking whether the assistant “remembers everything,” which does not tell you whether it can act correctly.

If the interface itself is failing, use the separate Claude availability guide to investigate that issue. Compacting a conversation organizes context; it does not establish the cause of a service error. More planning methods appear in the context engineering guides.

How to do this with fenbs

1. Create a free account, add the project, and create a task for the work that will continue.
2. Record the choices the person approved in Decisions and rules. Preserve their scope and identify unresolved questions.
3. Save the reviewed checkpoint in Where we left off, including the current sources, checks completed, and first next action.
4. Connect Claude or Claude Code using the appropriate integration guide. Ask the assistant to read the project context before it resumes.
5. Check the first continued result, then save updated progress so the next compact summary starts from the current state.

fenbs is not the right fit if all you need is Claude Code’s session compaction command. It does not compact Claude’s internal context or import chat history. It helps when the decisions and unfinished work need to remain available after the session changes.