More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Metric Diagnosis: Find Why a Number Moved

    1

    When a metric drops, agents list ten places to look or blame whatever changed that day. Your agent sizes the move, breaks it down, ranks causes, and passes three checks before it writes "root cause".

    Free

    Security scanned
    Metric Diagnosis: Find Why a Number Moved

    Metric Diagnosis: Find Why a Number Moved

    Example session with this skill installed

    Weekly checkout conversion fell from 3.1% to 2.4% last week. Figure out why. The events are in the warehouse table analytics.checkout_events.

    • Read your context and instructions
    • Compiled the metric diagnosis

    The agent first confirms the conversion formula and checks for a tracking change, then samples the top event types and user agents and finds nothing unusual. It compares the drop to normal weekly variance and last year, splits conversion into sessions and orders, and finds sessions grew while orders stayed flat. Segmenting by channel shows the new sessions came from one paid campaign with low intent. It ranks that above a same-day checkout release, which touched pages carrying under 10% of the lost orders, and reports: likely a healthy mix shift, not a regression, with the query that reproduces each step and one next check (conversion excluding the campaign).

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

    About this skill

    About this skill

    A metric move is a symptom, and a checklist of places to look is not a diagnosis. Asked why conversion fell, an agent will often name a change that shipped the same day, or build its case on a series it never inspected. Both answers read well and are wrong.

    This skill runs a seven-step investigation. The agent pins down the metric's formula and checks for a definition or pipeline change first, then samples the rows behind every series it uses. It sizes the move against normal week-over-week variance and last year, breaks the metric into a driver tree, and segments the branch that moved by time, region, internal changes, platform, competitors, cannibalization, and segment, cutting by outside provider first when outside platforms feed the metric. It ranks internal and external causes by fit to timing, shape, and segment, then confirms the top one before it classifies the move as a bug, a real regression, or a healthy drop.

    The proof

    The confirmation step comes from one metric-drop investigation where the skill had already fired and two wrong answers still shipped: one blamed a change made the same day that touched only a small share of the lost volume, and a later draft rested on a "clicks" series that mostly counted an internal API call. The skill now requires three checks before "root cause": the suspect's entities carried at least half of the lost volume, any claim about an outside system has direct evidence, and the agent runs any confirming query it can run itself instead of handing it off. There is no benchmark score; the evidence is these rules and the failure table at the end of SKILL.md, which maps each rule to the mistake it blocks.

    What you get

    One SKILL.md with the seven-step procedure, live-data and interview modes, the three confirmation checks, an output shape (verdict first, then a step-by-step table marked Verified or Unverified with a link to a reproducible query), and a table of 14 failure modes with the guard for each.

    How to use

    1. Add the skill folder to your agent's skills directory (for Claude Code, ~/.claude/skills/metric-diagnosis/).
    2. Ask why a metric moved, and give the metric name, the time window, and where the data lives.
    3. Review the verdict and the evidence table, then run the one next check the agent names to confirm or kill the top hypothesis.

    Source and license

    Free to use and MIT licensed. The full source, install steps, and issue tracker are on GitHub at https://github.com/robertmnyborg/analyst-rigor-skills.

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

    Does not connect to any data source itself. In interview mode it reasons without data and cannot verify anything. It expects you to keep an evidence store (queries and results) for the Verified label to mean anything. Group comparisons are handed to the separate causal-claim skill.

    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

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 8 days ago

    • Free with an Agensi account

    For live mode: query access to the metric's source data.

    Permissions

    No special permissions declared or detected

    Listed8 days ago

    What's inside

    Frequently Asked Questions

    Agents will list possible causes, and often stop there or pick the one that lines up by date. This skill forces a ranked lead hypothesis and blocks "root cause" until the suspect accounts for at least half the lost volume and the mechanism has direct evidence.

    For live mode, yes, through whatever query tools your agent already has. Without data it runs in interview mode, saying each step aloud and committing to a most-likely hypothesis.

    Yes. It is one SKILL.md file with no scripts or hooks, so any agent that reads SKILL.md can use it.

    The skill hands off to the separate causal-claim skill for that step. If you don't have it installed, the agent does the comparison without its pre-registration table.

    On GitHub at https://github.com/robertmnyborg/analyst-rigor-skills, under the MIT license. Every file in this package is there, plus the README and install steps.

    Free