More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Quick Strategy Backtest

    1

    Turn a trading idea stated in words into a runnable, honest backtest: precise specification, a look-ahead audit, realistic costs and market rules, and a report that shows drawdowns, not just returns.

    $5

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned
    Quick Strategy Backtest

    Quick Strategy Backtest

    Example session with this skill installed

    Backtest this idea: buy S&P 500 stocks that fall 5% or more in a single day, sell after 10 trading days. Use the last ten years of data.

    • Read your context and instructions
    • Compiled the quick strategy backtest

    SPEC (written before any code, for your correction):
    Universe: S&P 500 constituents point-in-time across the full ten years — including names later removed or delisted. Using today's member list would delete the worst losers and flatter the result.
    Signal: a stock's daily return is −5% or worse, computed from adjusted closes — known only after that day's close.
    Execution: buy at the NEXT trading day's open. Buying at the signal day's close is the classic look-ahead version of this idea; if you want same-day close execution, say so explicitly and the report will label it as assumed, not free.
    Hold/exit: sell at the open 10 trading days after entry. No stop-loss — you didn't specify one, and adding one after seeing results is tuning, not testing.
    Sizing: equal weight, max 20 open positions, sized from available cash; if more than 20 stocks signal on one day, take the 20 deepest drops and log the skips.
    Costs: commission plus 5 bps slippage per side, deducted from cash on every trade — with this turnover, costs are a first-order effect, not a footnote.
    Data: split/dividend-adjusted prices; names suspended on an entry or exit day are logged and marked at last real price, never filled at a fantasy price.

    WHAT GETS REPORTED: annualized return, volatility, Sharpe, max drawdown with its dates, win rate with average win/loss, trade count, turnover, and total costs as a share of gross P&L — plus the equity curve and drawdown chart. Because no parameters were tuned here (5% and 10 days came from you), no out-of-sample split is required; if we later explore other thresholds, the final two years get reserved and reported separately.

    KNOWN LIMIT BEFORE RUNNING: this is a falling-knife rule — it buys into news-driven drops as happily as into noise. The drawdown chart, not the headline return, is where that will show up.

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

    About this skill

    Most backtests are fiction with good charts — the strategy that "returned 40% a year" usually peeked at the future, ignored costs, or traded an index universe that quietly deleted its losers. This skill is the discipline for backtests you can trust, for daily-bar cash long positions in equities and ETFs. It starts with specification before code: universe (point-in-time, survivors and delisted alike), the signal and the moment it becomes known, which bar's price executes the trade, sizing, costs, and corporate-action handling — written in plain language using a fixed template, so you can correct the test before it runs. Then a look-ahead hunt walks every decision point — signal, selection, sizing, exit — asking whether that information existed at that moment, with a decision table converting common traps (same-bar execution on close-derived signals, full-sample rankings, fundamentals dated at quarter end instead of release) into honest redesigns. Accounting reflects reality: lot-level FIFO tracking, cash accounting that can't spend money it doesn't have, and market rules modeled where they apply — A-share T+1, lot sizes, price limits. Reporting covers the metrics that matter — annualized return, volatility, Sharpe, maximum drawdown with dates, win rate with average win/loss, turnover, and total costs as a share of gross P&L — with in-sample and out-of-sample results separated whenever parameters were tuned. Deliverables are a runnable script with frozen parameters, CSV/JSON exports, a dashboard (equity curve, drawdown, monthly returns, trade distribution), and a plain-language report with a limitations section. A lies-checklist and a 10-point report rubric gate delivery. The worked example runs a 60-day-high momentum rule on A-shares and reports the out-of-sample Sharpe decay instead of hiding it.

    • Daily bars and cash long positions only — intraday strategies, derivatives, shorting, and leverage are out of scope and declined rather than approximated.
    • A clean backtest still isn't a prediction: the report's limitations section names what the test cannot show, including regime change and capacity.
    • Point-in-time universe data matters — if only a current constituent list is available, the survivorship caveat is stated in the report rather than silently ignored.

    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 24 days ago

    Permissions

    No special permissions declared or detected

    Listed24 days ago
    Updated21 days ago

    What's inside

    Frequently Asked Questions

    Yes, with your explicit acceptance, and the report labels it as an assumption — a signal computed from the close can only execute at that close if you accept trading on the print itself. What the skill won't do is run it silently: that single choice is responsible for more fictional backtest returns than any other, so it gets named, not buried.

    Because today's S&P 500 list applied to 2016 deletes every company that failed or was removed since — exactly the stocks a dip-buying or momentum rule would have traded and lost on. If point-in-time data isn't available, the backtest can still run on today's list, but the report states the survivorship caveat up front so the number isn't mistaken for a clean result.

    Four things: the written specification (before code, so you can correct the test itself), a runnable script with frozen parameters, CSV/JSON exports of trades and the equity series, and a report with the dashboard charts, the full metric set including drawdown dates, and a plain-language limitations section.

    New in Trading & Investing