Counter-Hypothesis

    1

    The fastest way to lose a review is a causal claim nobody stress-tested.

    Free

    Security scanned

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    Counter-Hypothesis

    Example session with this skill installed

    counter-hypothesis: Weekly signups fell 18% because the onboarding redesign shipped on the 3rd.

    • Read your context and instructions
    • Compiled the counter-hypothesis

    Example on the fictional Ridgeline Outdoor shop.

    Claim: the redesign caused the 18% week-on-week drop in signups.

    1. 🔴

    Pipeline: the redesign renamed the signup event, and the dashboard still counts the old name. Test: compare new account rows with signup events for both weeks. Cost: S
    2. 🟡 Seasonality: the week after the summer sale always dips. Test: the same week last year. Cost: S
    3. 🟡

    Composition: paid traffic was cut the same week, and signup rate within each channel is flat. Test: signup rate by channel, before and after. Cost: M
    4. 🟢

    Selection: the comparison week was picked because it was a peak. Test: a 4-week baseline instead of 1. Cost: S
    5. 🟢

    Reverse causation: the team shipped the redesign because signups were already sliding. Test: the daily series for the 3 weeks before launch. Cost: S

    Cheapest disproof: count new account rows for both weeks. If they fell less than 5%, the drop is measurement and the claim is wrong.

    If still standing: signup rate by channel.

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

    About this skill

    The fastest way to lose a review is a causal claim nobody stress-tested. Reviewers ask for the metric, the baseline, the segment and the counter-hypothesis, and the fourth is the one most people skip. Counter-Hypothesis writes it before your reviewer has to.

    Give it any causal claim ("signups fell because of the redesign", "the lift came from the new banner") and it returns:

    • the claim restated in one line, so the counters target what you actually said
    • five alternative explanations, ranked by plausibility: composition shift, pipeline or measurement change, seasonality, reverse causation and selection bias. Each comes with a mechanism, a test that would prove your claim wrong, and a cost to run it.
    • the cheapest disproof: the one query to run first
    • an "if still standing" gate: what to test next if your claim survives

    Where it stops: it proposes tests, it doesn't prove causation. If the claim survives five strong counters, it says "Claim survives — proceed."

    Free and MIT-licensed. It's part of pm-os: https://github.com/hartejsingh99-cloud/pm-os

    It proposes tests; it doesn't prove causation, and it doesn't run the queries unless your agent already has a database tool. The counters are only as sharp as the context you give it.

    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

    Free forever

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 5 days ago

    • Free with an Agensi account

    Permissions

    Read Files

    File Scopes

    .pm-os/house-rules.md
    ~/.claude/pm-os/house-rules.md

    Reads your optional house-rules file, plus any notes or data you point it to. It never writes files or calls the network.

    Listed5 days ago

    What's inside

    Frequently Asked Questions

    Yes. It's free and MIT-licensed, and the source is on GitHub: https://github.com/hartejsingh99-cloud/pm-os

    Not by itself. It names the cheapest test and the threshold that would kill the claim. If your agent already has a database tool, it can run it.

    No. With one (start from references/house-rules.example.md in the skill), it matches your reviewer's format and rules out counters your known data traps already explain.

    New in Business & Operations