Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Task Orchestrator: Sub-Agents Only When They Pay Off

    1

    Sub-agents are easy to start and expensive to get wrong.

    $6.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Task Orchestrator: Sub-Agents Only When They Pay Off

    Task Orchestrator: Sub-Agents Only When They Pay Off

    Example session with this skill installed

    Migrate our Express API from JavaScript to TypeScript. There are 40 route files, a shared db layer and Jest tests. Do it in parallel where it's safe, but don't break the build.

    • Read your context and instructions
    • Compiled the task orchestrator

    Run: express-ts-migration (20261112-140512)

    Size: HIGH (40 files, shared types, build at risk)
    Stage: Implementation, as requested. Deploy and publish are out of scope.

    Plan

    StreamOwns (write)Reads onlyModel / effortDepends on
    A. Foundationstsconfig.json, src/types/**, src/db/**everythingstrongest / high (shared types every other stream depends on)none
    B. Routes 1-20src/routes/a-m/**src/types/**, src/db/**mid / mediumA
    C. Routes 21-40src/routes/n-z/**src/types/**, src/db/**mid / mediumA
    D. Teststest/**, jest.config.tssrc/**mid / mediumB, C

    Why not more parallelism: the db layer and shared types must exist first, so A runs alone; B and C have disjoint folders and run together.

    Checkpoints

    State saved to .cursor/orchestration/20261112-140512-express-ts-migration/STATE.json after every wave. If the run stops, the next session resumes from the last verified stream.

    Results

    • A: 9 files converted, tsc --noEmit clean on foundations
    • B: 20 routes converted, 3 any left with TODO and reason
    • C: 20 routes converted, 1 route needed a db type change, returned to A instead of editing src/db itself
    • D: Jest switched to ts-jest, 148 tests

    Verification (main agent)

    CheckResult
    tsc --noEmit on whole projectPASS
    npm testPASS, 148/148
    npm run buildPASS
    Manual smoke of 3 endpointsNot verified (no running database in this environment)

    Next: start the API against a dev database and hit /health, /users, /orders before merging.

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

    About this skill

    Sub-agents are easy to start and expensive to get wrong. Two agents edit the same file. A huge model is used to rename a variable. A long run dies halfway and nobody knows what finished. The final message says "done" and nothing was tested.

    The skill gives your agent one way of working that scales from a one-line question to a multi-day refactor.

    What your agent does

    • Sizes up every task first. Intent, the stage you asked for (review, plan or build), what must not change, and how it will prove the work. Each task and subtask gets a size: trivial, low, medium, high or critical.
    • Stays direct when it should. Trivial and low tasks never spawn agents, scan tools or create planning files.
    • Delegates only when it's worth it. Two or more ready workstreams, write sets that don't overlap, and a benefit larger than the coordination cost. Otherwise one agent does it.
    • Picks the least model that will do. Each sub-agent gets the cheapest model and effort level that fits its job, and the skill records what it asked for against what it actually got.
    • Writes precise briefs. Every worker gets a context packet with its goal, files it owns, files it may only read, rules, and the evidence it must return.
    • Survives interruptions. Run state is saved to disk with a schema, so a crashed or paused run picks up from what the repository actually shows.
    • Integrates and verifies. The main agent merges results, runs the checks and reports PASS, FAIL or "Not verified". Nothing is called done without evidence.

    Respects your boundaries

    Review-only stays review-only. A plan stays a plan until you approve it. Sub-agents never deploy, publish, delete data or change permissions on their own.

    What's inside

    • SKILL.md with the full operating policy
    • 7 reference guides (request refinement, resource routing, contracts, state and resume, host integration for Claude Code, Codex, Cursor and Gemini/Antigravity, acceptance tests, a full demonstration) and 2 worked examples
    • Plan, run, context and evidence templates plus a JSON schema for run state
    • 2 Python helpers (state store and resource discovery) with 33 unit tests

    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.

    Parallel work needs an agent that supports sub-agents; otherwise it runs the same plan one step at a time. Saved run state and resume need Python 3.9+ with the jsonschema package. Model routing only uses the model and effort controls your agent actually exposes.

    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 day ago

    Software
    An AI coding agent with file accessPython
    Optional
    Python 3.9+ with jsonschemaGit

    Permissions

    Terminal / Shell
    Read Files
    Write Files
    Network Access
    Environment Variables
    Listed1 day ago

    What's inside

    Frequently Asked Questions

    No. Trivial and low tasks are handled directly with no planning files and no sub-agents.

    It routes each sub-task to the least sufficient model and effort level your agent exposes, and records requested versus actual. It never claims savings it can't show.

    The run state is saved to disk with a schema. The next session reads it, checks the repository, and resumes from the last verified step.

    New in Agents & Orchestration