Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Model Router Intermediate: Delegate Across Your AI CLIs

    1

    One main agent routes work across Claude Code, Codex, Grok and other CLIs you pay for, with scoped briefs, a bundled runner that checks every answer, and independent review.

    $14.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Model Router Intermediate: Delegate Across Your AI CLIs

    Model Router Intermediate: Delegate Across Your AI CLIs

    Example session with this skill installed

    I just downloaded the model-router-intermediate skill into ./skills/model-router-intermediate. Read its SKILL.md and follow it.

    My setup: Claude Code is my main agent (Opus, Sonnet, Haiku on a Claude Max subscription). I also have the Codex CLI (ChatGPT Plus subscription) and the Grok CLI installed and signed in.

    Do the first-run setup for this project now (write the filled files into this project as the skill says), including enabling the right lanes and running the runner doctor. Do not run --doctor --run. Then show me how you would split this task across my tools: "add a password reset flow to our Express API, with email tokens and rate limiting, and get it reviewed before release".

    End with a short summary of what you set up and the routing plan.

    • Read your context and instructions
    • Compiled the model router intermediate

    (Real run, 2026-10-03: Codex CLI as the agent, the intermediate pack unzipped into a fresh Express project. Final reply, unedited apart from local file links. The run prompt also told the agent to ignore unrelated instructions on the test machine.)

    Setup written to ai-orchestrator/.

    Configured: Filled rules, briefs and protocols; enabled claude, codex, grok; kept defaults unpinned; added the routing import to CLAUDE.md.

    • Passed: Presence-only doctor, exit 0, plus static checks. Did not run --doctor --run.

    Limits: Claude CLI reports no login; Grok’s model refresh failed. Model execution, quota and fresh-session rule loading remain unverified.

    Proposed password-reset split

    ToolResponsibility
    Claude OpusPlan security contracts, implement token lifecycle and rate limiting, coordinate and merge
    Claude SonnetEmail integration and tests in assigned files
    Claude HaikuBounded source inventory
    GrokConditional primary-source research and companion scope check after access/billing verification
    Codex Astra, xhighOne read-only security audit of the merged diff

    Before release: reproduce findings, fix with regressions, replay acceptance checks, and resolve pending review coverage.

    Summary: Project routing is configured and doctor passes; Claude builds and merges, Codex independently audits, and Grok supplies conditional research and coverage review.

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

    About this skill

    The problem: you pay for several AI tools, but one of them does all the work while the others sit idle. When you do hand work off, the other CLI gets a vague prompt, exits "successfully" with nothing, and nobody notices.

    What you get: everything in the Beginner pack, plus a delegation method and a runner. Your main agent keeps coordinating, decides which of your tools fits each job, hands it a scoped brief, and checks the result before accepting it.

    What your agent learns

    • Lanes: each extra AI tool becomes a lane your main agent can call with a task brief, chosen by capability, context size, permissions and billing (local, free, subscription, then pay-per-token).
    • Routing rules: a multi-tool decision tree, a delegation matrix with the reason for every assignment, and model tiers with effort guidance.
    • Independent review: reviews go to a different model family from the one that wrote the code; a same-family check is labelled as a self-check.
    • Research triage: independent questions go to tools with live access, and claims are checked against primary sources before they are used.
    • Split builds: every worker gets the whole scope and its own section; one owner merges and names conflicts before the single audit.

    The bundled runner

    bin/cli-run.mjs (Node 18+) calls the CLI for a lane, reads its final result, and exits non-zero when the answer is missing, interrupted, timed out or the wrong shape. It tells authentication, quota and missing-tool failures apart, and --expect-file / --expect-json enforce an output contract. Supported lanes: Claude Code, Codex, Antigravity (Gemini), Grok, Hermes and Qwen. Every lane starts off; setup turns on only the CLIs you actually have.

    How setup works

    On first run your agent asks which AI tools you have, copies the rules and runner into your project, fills in your stack table, enables your installed CLIs, and runs the runner's doctor to confirm they are present. No installer and no account.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.

    Requires Node.js 18+, pre-installed and authenticated worker CLIs. Does not handle credential management. Doctor command must pass for each lane before execution.

    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 buying a skill.

    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 1 month ago

    Software
    Node.jsUse Claude Code As An MCP

    Permissions

    Terminal / Shell
    Read Files
    Write Files

    The runner starts the worker CLIs you enable, keeps a local log at ~/.ai-orchestrator/cli-run.log.jsonl, and makes no network calls of its own; each CLI uses its own sign-in.

    Listed1 month ago
    Updated1 month ago

    What's inside

    Frequently Asked Questions

    This skill enables your primary AI agent to intelligently distribute tasks across multiple CLI tools like Claude Code, Grok, and Codex, ensuring you maximize the value of your various subscriptions while maintaining high quality through automated result verification.

    The bundle includes a decision matrix for routing tasks, a delegation framework for providing scoped briefs, and the 'cli-run.mjs' runner which validates that sub-agents actually produced the requested output before finishing.

    The skill currently supports Claude Code, Codex, Antigravity (Gemini), Grok, Hermes, and Qwen, but it only activates the specific lanes for the tools you already have installed and configured.

    The bundled runner uses an '--expect' contract to detect empty responses, timeouts, or malformed data, and the system is designed to route code reviews to a different model family than the one that authored the code to ensure objective oversight.

    Setup is handled by your agent, which will walk you through a 'doctor' check to identify your installed tools and automatically configure your stack table without requiring an external installer or new accounts.

    New in Agents & Orchestration