Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Save Before Compact

    1

    Say "save before I compact" (or "checkpoint this session") and the agent: 1.

    Free

    Security scanned

    About this skill

    Compaction keeps a summary of the conversation and drops detail. The things most likely to vanish are the ones decided after the main work was written up, and questions the agent asked you that you never answered.

    Say "save before I compact" (or "checkpoint this session") and the agent:

    1. Writes a one-line RESUME.md stub first, before scanning, because the scan itself costs context and you usually run this when context is nearly full.
    2. Sweeps the chat for decisions and reversals, verified facts that change the plan, feedback on how to work, project state and next actions, anything you said to remember, and questions it asked you that are still open.
    3. Routes each item: cross-session facts to whatever memory system you have (Claude Code's auto memory, a memory MCP, or plain note files), project decisions to the project file they belong in.
    4. Overwrites RESUME.md with one short dated block that a session with zero memory of this chat can act on.
    5. Only then reports in chat: what it saved, what it left unsaved on purpose, or "nothing new to save, you're safe to compact".

    The order is deliberate. In an audit of my own transcripts, 5 of the 6 compactions that fired while a save like this was running ate the closing chat text, while the file writes had already landed. So files go first and chat text last. The numbers and their caveats are in EVIDENCE.md, which ships with the skill.

    It is not a memory system and it doesn't run on its own. Works with catnap, a free Claude Code plugin that lets Claude compact itself at a clean break. Not affiliated with Anthropic.

    One person, one Windows machine, one Claude Code version behind the evidence. The 5-of-6 figure is six compactions picked because the skill was running when they fired; it is not a compaction loss rate. It hasn't been compared against a plain "save what matters before I compact" prompt.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 9 days ago

    • Free with an Agensi account

    Permissions

    Read Files
    Write Files
    Listed9 days ago

    What's inside

    Frequently Asked Questions

    If you only ever compact mid-task and never clear, it may be. A steered summary is still a summary you don't control, and it's gone the moment you /clear or close the terminal. Files survive that.

    Two runs, one has to come back red. In a fresh session with two trivial turns, it must answer "nothing new to save" and write nothing. Then state a decision ("actually, go with option B") and run it again: that decision must show up in RESUME.md.

    New in Productivity

    Free