Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Reduce AI Token Router

    1

    Keep the main chat as the supervisor. Send routine work to a fast model, and step up to a stronger model only when that same task fails. Works in Claude Code, Codex, Gemini, and Grok.

    $9.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Reduce AI Token Router

    Reduce AI Token Router

    Example session with this skill installed

    Task name: product card price
    Add the price from lib/api.ts onto components/ProductCard.tsx.
    Done when the card shows the price returned by getProduct.

    • Read your context and instructions
    • Compiled the reduce ai token

    Summary

    Changed components/ProductCard.tsx. The card now shows the price returned by getProduct. No other file was edited.

    PieceResult
    Filecomponents/ProductCard.tsx
    First trySonnet
    CheckThe card shows getProduct().price

    What the child reported

    The price comes from lib/api.ts. ProductCard now reads getProduct().price and renders it under the product name. The page in app/ was left alone.

    If the price is wrong

    Send the same task again

    retry product card price
    The price renders as undefined. Fix only components/ProductCard.tsx.

    The second try stays on Sonnet at higher effort. Later tries step one model up. The last model fixes that line. The page is not rebuilt.

    Review

    After the card exists, the reviewer opens with one line: ship, fix, or redo. It does not edit the file.

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

    About this skill

    The expensive part of a coding chat is the history. Every turn sends the conversation again, so a long main chat keeps paying for work that already happened.

    This skill makes the main chat the supervisor. It names one task, sends a short brief, and reads the summary. The child does the edit. The first try uses a fast model. The second try raises effort on that same model. Later tries step one live model up.

    Claude climbs Haiku, Sonnet, Opus, then Fable. Codex climbs Luna, Terra, Sol, then Astra. Gemini climbs Flash-Lite, Flash, then Pro. Grok climbs Grok Build, then 4.3, 4.5, 4.6, and 4.7. The last model fixes only the part that failed. The whole task is not rebuilt.

    People on the same repo can use different apps. Each app keeps its own retry count. A few lines at the top of the project rules tell the model what the app is, so it does not re-read the repo to learn the flow.

    What the buyer gets:

    • A skill the agent loads before it delegates or retries
    • The job of each model, in reference.md
    • A shop-app walkthrough, in examples.md

    The agent follows these instructions in the chat. It does not install scripts into your project. You still use the models your app already has. The last failed part is fixed in the main chat.

    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

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 2 days ago

    Permissions

    Read Files
    Write Files
    Listed2 days ago

    What's inside

    Frequently Asked Questions

    The main chat stays short. Routine work goes to a fast model first. The same task steps up to a stronger model only after that try fails, so the expensive model is used for the part that needs it.

    Claude Code, Codex, Gemini CLI, and Grok. Your agent loads the skill through Agensi and names the model for each try. The first try is Sonnet, Luna, Flash-Lite, or Grok Build, depending on the app.

    No. Connect your agent to Agensi, sign in, and ask it to use Reduce AI Token Router. There is no Python script to download or run.

    Three files: SKILL.md, reference.md, and examples.md. SKILL.md is the rule your agent follows. reference.md is the job of each model. examples.md is a shop-app walkthrough. Your agent loads them when you ask. They are not saved into your project.

    The child cannot see the main chat. Each try gets a short brief that names the file, the change, and how to tell it is done. A retry keeps the same task name and fixes only the part that failed. The main chat reads the summary.

    New in Agents & Orchestration