Growth experiment design
Skill duandigi/duandigi-growth-marketing-skill/skills/growth-experiment-design
Evidence-first AI Agent Skills for growth marketing — multi-channel analytics (SEO, paid media, social, local search, CRM), secure account integration, AI evaluation, experimentation & approval-safe optimization for Claude Code.
npx -y skills add duandigi/duandigi-growth-marketing-skill --skill growth-experiment-designAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 26 days oldThe repository was created 26 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Use this skill when turning a growth opportunity into a testable experiment with a falsifiable hypothesis, intervention, audience, primary metric, guardrail metrics, tracking plan, decision rule, and rollback plan.
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
2.8 KB, 455 tokens by cl100k_base, as published. Nobody here has run it
Growth Experiment Design
Purpose
Design experiments that can produce a trustworthy decision, not merely activity.
Inputs
- Opportunity and supporting evidence
- Baseline metric and target audience
- Implementation constraints
- Available tracking and expected traffic or sample
If critical input is unavailable, label it unknown and create a research or instrumentation task. Do not invent values.
Workflow
- Write a falsifiable hypothesis in the form: because evidence suggests X, changing Y for audience Z should move metric M through mechanism K.
- Choose the smallest intervention that can test the mechanism.
- Define eligibility, exposure, variants, exclusions, and contamination risks.
- Choose one primary metric and a small set of guardrail metrics.
- Define the measurement window, minimum data requirement, and decision rule before launch.
- Specify instrumentation, QA, owner, launch checklist, rollback trigger, and stopping conditions.
- If a controlled test is impossible, design a time-boxed observational test and label its weaker causal confidence.
Required output
Return a concise, decision-oriented response containing:
- Hypothesis
- Experiment design
- Audience and variants
- Metrics
- Tracking plan
- Decision rule
- Risks and rollback
Label important statements as confirmed, inferred, assumed, or unknown when the distinction affects the decision.
Guardrails
Do not:
-
Changing multiple unrelated mechanisms in one test
-
Selecting several primary metrics
-
Promising statistical significance without adequate data
-
Claim guaranteed growth or present an estimate as observed fact.
-
Recommend spam, fake reviews, impersonation, deceptive urgency, dark patterns, policy evasion, or unauthorized production changes.
-
Hide material uncertainty, tracking limitations, or possible harm.
When an action can spend money, publish content, contact people, change production systems, delete data, or alter access, produce a plan and request explicit authorization rather than executing automatically.
Completion check
Before finishing, verify that the output:
- answers a specific growth decision;
- uses the supplied business context;
- separates evidence from assumptions;
- defines a measurable next step;
- includes risks, constraints, and missing data;
- is no longer than necessary for the decision.
Gives 0 of the 12 instructions most analytics metrics skills give in 455 tokens
Counted across 368 of the 369 authors here whose files we hold, read 2026-08-06
- read product marketing context before asking questionsin 18 of 368, across 12 files
- use lowercase with underscores for event namesin 16 of 368, across 6 files
- track events for decisions not vanity metricsin 15 of 368, across 5 files
- use object-action format for event namesin 15 of 368, across 8 files
- produce a tracking plan documentin 14 of 368, across 4 files
- Call RUBE_SEARCH_TOOLS first to get current schemasin 13 of 368, across 2 files
- establish consistent event naming conventions before implementingin 10 of 368, across 4 files
- Verify dimension and metric compatibility before reportingin 9 of 368, across 2 files
- Encrypt data at rest and in transitin 9 of 368, across 3 files
- use snake_case for event namesin 9 of 368, across 5 files
- monitor technical health during the testin 9 of 368, across 5 files
- use consistent property namesin 8 of 368, across 4 files
Said here and by no other author read
- Choose the smallest intervention testing the mechanism
- Define eligibility, exposure, variants, exclusions, and contamination risks
- Define measurement window, data requirement, and decision rule before launch
- Specify instrumentation, QA, owner, launch checklist, rollback trigger, and stopping conditions
- Design an observational test if controlled testing is impossible
- Label missing critical inputs as unknown
Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.