Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Subagent-Advisor

    1

    Calculate if subagent delegation saves time or just wastes money using calibrated token and cost estimates.

    $9

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Subagent-Advisor

    Subagent-Advisor

    Example session with this skill installed

    I have 16 large features to implement. All can run in parallel. I need to read 300,000 tokens of docs. I'm using Opus 5.5. Should I use subagents?

    • Read your context and instructions
    • Compiled the subagent-advisor

    I have calculated the trade-offs for delegating these 16 features versus handling them in the main session.

    🟢 Use subagents

    16 steps · 16 in parallel · large · Opus 5.5

    16 steps run at the same time: finishes 79% sooner, at 191% more cost.

    COST     alone      ███████░░░░░░░░░░░░░     $6.21
             subagents  ████████████████████    $18.08   +191%
    
    TIME     alone      ████████████████████    17 min
             subagents  ████░░░░░░░░░░░░░░░░     4 min   −79%
    
    TOKENS   alone      ██████████████░░░░░░      9.2M
             subagents  ████████████████████     13.0M   +41%
    

    Typical values; the real figure can reach 3×.

    → I recommend splitting into 16 subagents. Should I use subagents? (yes / no)

    Next steps

    1. Type yes to trigger the delegation of the 16 parallel workers.
    2. Monitor the first wave of output to ensure subagent read performance matches the estimate.
    3. Review the summary report once all workers return their results.

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

    About this skill

    Developer agents often hallucinate the efficiency of parallel execution. You might think splitting a task into five subagents will save time, but the token overhead from cold starts and context duplication can increase your bill by 10x without delivering a faster result. This skill provides a mathematical framework to stop guessing and start calculating the real cost of delegation.

    What it does

    • Multi-step Analysis evaluates whether a task should be handled by the main session or split across parallel workers.
    • Cost Estimation calculates projected spend in USD based on specific model pricing like Sonnet 3.5, Opus, or Haiku.
    • Time Benchmarking compares sequential processing seconds against parallel wave execution.
    • Context Forecasting predicts token growth to prevent hitting context window caps prematurely.
    • Decision Verdicts provides a "use", "maybe", or "skip" recommendation based on calibrated transcript data.

    How it works

    1. Input Task Metrics by defining the number of steps, independent threads, and tokens to be read from files or docs.
    2. Apply Calibration Constants using built-in stats for worker startup costs (69k tokens) and reread overlaps.
    3. Calculate Parallel Waves to determine how many sequential "rounds" of execution are actually required.
    4. Compare Plans side-by-side using standardized USD/token rates for the chosen model.
    5. Review Verdict via a visual bar chart comparison before confirming the delegation plan.

    Frameworks & tools

    This skill is model-agnostic but includes pre-calibrated rates for various model families including sonnet, opus, haiku, and fable/mythos. It is designed for use within agentic environments like Claude Code or Cursor where subagent delegation is a native capability.

    Why this beats prompting it yourself

    Generic prompts fail to account for the massive 69k+ token "startup tax" every subagent incurs. This skill uses specific coefficients derived from real-world telemetry to ensure estimates are within 15% of actual spend on average, rather than just guessing.

    Use cases

    • Complex Refactoring determining if 20 file edits should happen in one session or five workers.
    • Large-Scale Research calculating the cost of reading 300k tokens of documentation across parallel threads.
    • Budget Management preventing runaway costs on expensive models like Opus by identifying sequential-only tasks.
    • Performance Optimization identifying when the "startup tax" of subagents exceeds the time saved by parallelizing small tasks.

    Known limitations

    The current calibration assumes subagents start with ~69k tokens of system prompt and tool definitions. Accuracy for a single task can vary by ±80 percent, so it always presents a range.

    Calibrated for agents with ~69k token startup overhead. Estimates for single tasks may vary by ±80%. Accuracy depends on model-specific pricing accuracy.

    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 8 days ago

    Permissions

    Read Files
    Listed8 days ago

    What's inside

    Frequently Asked Questions

    The Subagent Advisor accounts for the "startup tax"—the massive token overhead (approx. 69k tokens) incurred every time a subagent is initialized—preventing expensive bill spikes that simple prompts usually ignore.

    The skill is model-agnostic and works with any agentic environment like Claude Code, Cursor, or custom frameworks, provided you use the pre-calibrated rates for models like Sonnet, Opus, or Haiku.

    No complex integration is required; simply input your task metrics (steps, threads, and document size) into the advisor to receive a mathematical breakdown and a 'Use' or 'Skip' recommendation.

    You receive the full mathematical framework, calibrated coefficients for current top-tier models, and the logic required to generate side-by-side USD and time-benchmarking comparisons.

    Due to the 'startup tax' and variable LLM outputs, single-task accuracy typically falls within ±15% of actual spend, though individual variance can reach ±80%, which is why the tool provides a range rather than a single number.

    The skill includes built-in stats for common model families, and the developer provides periodic calibration updates to reflect changes in model pricing and system prompt overheads.

    New in Agents & Orchestration