More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Smart Agent Router: Pick the Right Model, Cut the Cost

    1

    One command sets up Haiku, Sonnet and Opus subagents with least-privilege tools, then shows you what the routing saved in a visual dashboard.

    $3

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Smart Agent Router: Pick the Right Model, Cut the Cost

    Smart Agent Router: Pick the Right Model, Cut the Cost

    Example session with this skill installed

    /agent-router report

    • Read your context and instructions
    • Compiled the smart agent router

    SAVED BY ROUTING TO SUBAGENTS $156.74 subagents handled 50% of tokens
    AT THIS PACE $111.96 saved per month

    Whole project cost
    actual $243.22 39% less
    all on Opus 5.5 $399.96

    Cost by model (actual / same tokens on Opus 5.5)
    Haiku 5.5 sub $3.37 / $120.67 saved $117.30
    Sonnet 5.5 sub $50.21 / $100.42 saved $50.21

    (Sample data from --demo. Your figures come from your own usage.)

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

    About this skill

    Smart Agent Router

    Stop spending Opus-level prices on simple tasks.

    Smart Agent Router configures Claude Code to delegate work to the right model based on task complexity. Set it up once, run /agent-router, and track your savings with a built-in dashboard.

    No API proxy. No background service. Nothing to host or configure with extra API keys. It uses Claude Code's native subagents.

    What You Get

    1. Smart Model Routing

    Run /agent-router in any project. It:

    • Detects your project type: Code, documents, finance, media, web design or research.
    • Proposes a plan first: Nothing changes until you approve. Existing files are never silently overwritten.
    • Routes tasks by complexity:
      • T1 – Haiku: Search, file lookups, logs and data extraction.
      • T2 – Sonnet: Routine edits, scripts and media conversion.
      • T3 – Opus: Complex debugging, analysis and architecture reviews.
      • T4 – Fable: Frontier-model escalation, only with your approval.
    • Keeps Opus 5.5 as orchestrator and delegates suitable work to subagents.
    • Adds task-specific checks: Tests and linters for code, totals checks for finance, and ffprobe for media.
    • Maintains project memory: Updates a marked routing section in CLAUDE.md and creates lean docs/ memory for future sessions.

    2. Savings Dashboard

    Run /agent-router report whenever you want to see how much you've saved.

    The report uses your project's local Claude Code history to show:

    • Total savings compared with running the same tokens on Opus 5.5.
    • Running savings chart and monthly pace (after one week of history).
    • Cost by model versus the all-Opus baseline.
    • Token usage by tier, tasks per subagent and daily savings.
    • Plain-language insights on your biggest savings and what to route next.

    Choose your format:

    • HTML: A self-contained dashboard with light and dark themes, designed for desktop and mobile.
    • Terminal: The same report in your command line.
    • JSON: Use --json to integrate with your own tools.
    • Demo: Use --demo to preview the dashboard with sample data.

    Built with Safety in Mind

    • Subagents receive only the tools they need. Read-only agents cannot edit; writers are restricted to the output folder.
    • File, document and log contents are treated as data, not instructions, helping guard against prompt injection.
    • Optional permissions.deny rules can block access to .env files, keys and secrets/.
    • Agents cannot push, deploy, install or access the network unless explicitly instructed.
    • The skill refuses to run in home, drive-root or system folders.
    • The included Python script is read-only, uses the standard library and has no network access. It never prints prompts or file contents.

    What's Included

    • SKILL.md – Setup and routing workflow.
    • routing_report.py – Savings report and dashboard.
    • 8 subagent templates and 3 project-memory templates.
    • CLAUDE.md routing section template.
    • README.md, SECURITY.md, CHANGELOG.md and LICENSE.md.

    Licence

    One purchase covers one person, team or company (one organization).

    Private-repository use is allowed. Resale and public redistribution are prohibited.

    Undo Anytime

    Every change is documented. Follow the README to remove the setup in under a minute.

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for LLM Ops.

    Savings are estimates at list API prices; on a subscription plan they show relative usage, not money billed. It configures routing inside Claude Code only; it is not an API proxy and does not route other tools. Routing rules are instructions the model follows; the enforced controls are the tool lists and deny rules. The report covers one project at a time and needs Python 3.9 or later.

    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 4 days ago

    Software
    Claude CodeOpus, Sonnet and Haiku accessGit
    Optional
    Python 3.9+ (for the report)
    FFmpeg
    orImageMagick (media work)

    Works on Windows, macOS and Linux. Claude Code v2.1.233 or later is recommended. The report needs no installs; it uses only the Python standard library.

    Permissions

    Terminal / Shell

    File Scopes

    project
    folder:
    .claude/agents/
    .claude/settings.json
    CLAUDE.md
    docs/
    ~/.claude/projects/
    (read-only
    for
    the
    report)

    Nothing is written until you approve the plan. The report is read-only and offline, and prints only model names, token counts, subagent names and cost estimates.

    Listed5 days ago

    What's inside

    Frequently Asked Questions

    No. The setup is Markdown templates, and the report runs offline on your machine. The package contains no URLs.

    Not without asking. You see the full plan first and choose overwrite, merge or skip for each existing file.

    The report re-prices the tokens each model used at Opus 5.5 list prices and shows the difference. Escalations to Fable cost more by design and are shown apart.

    No. It uses the models your Claude Code account already has.

    Yes. The dollar figures are estimates, but the tier mix and findings show how much work left your most limited model.

    Yes. Ask for the demo, and it renders with built-in sample data labelled as such.

    Yes. It detects documents and finance, media, web design and research work and creates only the agents that folder needs.

    Yes. One purchase covers one organization: a person, or a team or company. You may commit the skill to your private repositories. See LICENSE.md.

    The final report lists every file it created or changed, and the README has removal steps.

    New in LLM Ops