Conversation Split Advisor

    1

    Detects topic drift and tells you whether to start a fresh session, with a ready-to-paste carry-forward summary.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Conversation Split Advisor

    Example session with this skill installed

    This session has covered a lot. /suggest-split — should I start a new conversation?

    • Read your context and instructions
    • Compiled the conversation split advisor
    ✂️ Split recommended — conversation has drifted across 3 topic(s)
    
    Topic map:
      Turns 1–8    Debugging login timeout           🗑️ Discard — resolved, no longer relevant
      Turns 9–14   Reviewing PR #42                  📋 Summarize — 2 decisions still apply
      Turns 15–now Refactoring auth middleware        ✅ Current task — carry forward
    
    What to carry into the new conversation:
    --- CONTEXT FROM PREVIOUS SESSION ---
    - Decided to use JWT over sessions (PR #42 review, turn 11) — shapes the current refactor
    - auth.ts middleware must stay stateless — agreed in turn 13
    - Current task: refactor auth middleware in src/middleware/auth.ts
    --- END CONTEXT ---
    
    Split instructions:
    1. Copy the context block above
    2. Start a new Claude Code session: claude
    3. Paste the context block as your first message
    4. Continue with: "Continue refactoring src/middleware/auth.ts"
    

    When the session is still focused, it says so:

    ✅ No split needed — conversation is focused on a single task (12 turns, ~24,000 tokens used)
    Context is coherent. Continue in this session.
    

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Maps the conversation into distinct topic blocks, scores each for carry-forward value relative to your current task, and recommends whether starting a new session would meaningfully improve context quality. If yes, produces a ready-to-paste context block for the new session.

    When to use

    • A session has been running for a long time across many different tasks
    • You feel like your agent has "lost the thread" of what you're working on
    • Context is over 70% full and you're not sure how much is still relevant

    How it works

    Invoke it by name in your agent (in Claude Code: /suggest-split). It reads only the current conversation — no files are written, no network calls, no external services. You decide whether to start the new session.

    Part of Claude Token Inspector, an open-source (MIT) suite of 8 context-window skills by the original author. Source: https://github.com/msishariful/claude-token-inspector

    Requires at least 5 turns to evaluate. Does not recommend splits for sessions under 10 turns unless topic drift is extreme.

    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 23 days ago

    • Free with an Agensi account

    Permissions

    No special permissions declared or detected

    Listed23 days ago

    What's inside

    Frequently Asked Questions

    As sessions grow, irrelevant data consumes the 'context window,' causing the AI to slow down or forget earlier constraints. This skill proactively identifies when your current chat has become inefficient and prepares a condensed summary to keep your next session fast and focused.

    It's a standard SKILL.md skill, built and tested on Claude Code. It also works in other SKILL.md-compatible agents (Cursor, Codex CLI, Gemini CLI, etc.); the carry-forward block is plain text you can paste into any new session.

    The SKILL.md instructions file, which includes the topic-mapping steps, split thresholds and output templates. It's free and open source (MIT) and part of the Claude Token Inspector repo on GitHub.

    No, 'suggest split' is designed to be lightweight and local. It analyzes the conversation history already present in your current session buffer to provide its recommendations without requiring external API calls or database storage.

    Not through a setting. The thresholds are written into the SKILL.md (e.g. split when over 40% of tokens come from unrelated turns, or context is over 70% full with dead-weight early turns). Since it's open source, you can edit the file to change them.

    New in Agents & Orchestration