experiment-design-kit — Claude Code skill: what it does
experiment-design-kit structures product and growth hypotheses into test briefs with guardrails, variant catalogs, and rollback plans.
Apache-2.0 ★ 450 updated 2026-04-03
npx skills add gtmagents/gtm-agents --skill experiment-design-kit -a claude-codeWhat experiment-design-kit does
This skill guides the preparation of growth and product tests using a five-step structure. It starts by defining the user problem and primary metric, then shapes the hypothesis using a structured persona-action-outcome model. The workflow continues into sample sizing, variant rules, and operational execution.
Included templates provide experiment briefs, variant matrices across messaging surfaces, rollback playbooks, and backlog boards scored via ICE or RICE. It also includes an outline for weekly stakeholder packets covering KPI trends and pipeline impact.
Before authorizing a build, the workflow requires predefined KPI guardrail thresholds. For instance, activation rate changes must remain within ±3% of baseline, while revenue drops per visitor cannot exceed 2% across 48 hours.
What it helps with
- frame testable hypotheses with primary metrics and power calculations
- build variant matrices for segmented messaging and user states
- establish metric guardrails and emergency rollback procedures
- generate weekly experiment reports for cross-functional review
Example prompts for the agent
- Draft an experiment brief for our checkout page redesign using experiment-design-kit
- Build a variant matrix and rollback playbook for the new pricing experiment
- Create a Weekly Experiment Packet outline summarizing this week's active tests
How to install experiment-design-kit
1Skills CLI (Vercel)
npx skills add gtmagents/gtm-agents --skill experiment-design-kit -a claude-code2Manually: put the files in place
~/.claude/skills/experiment-design-kit/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/experiment-design-kit/
1Skills CLI (Vercel)
npx skills add gtmagents/gtm-agents --skill experiment-design-kit -a codex2Manually: put the files in place
.agents/skills/experiment-design-kit/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add gtmagents/gtm-agents --skill experiment-design-kit -a cursor2Manually: put the files in place
.cursor/skills/experiment-design-kit/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install gtmagents/gtm-agents plugins/growth-experiments/skills/experiment-design-kit2Manually: put the files in place
.github/skills/experiment-design-kit/SKILL.md1Skills CLI (Vercel)
npx skills add gtmagents/gtm-agents --skill experiment-design-kit -a gemini-cli2Platform CLI
gemini skills install https://github.com/gtmagents/gtm-agents --path plugins/growth-experiments/skills/experiment-design-kitManually: put the files in place
.devin/skills/experiment-design-kit/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about experiment-design-kit
How to install experiment-design-kit in Claude Code?
Run npx skills add gtmagents/gtm-agents --skill experiment-design-kit -a claude-code in your terminal. The skill file is placed at ~/.claude/skills/experiment-design-kit/SKILL.md.
What does experiment-design-kit do?
It standardizes experiment design by organizing problem framing, hypothesis syntax, measurement criteria, and guardrail alerting into repeatable templates.
What default guardrail thresholds does the kit include?
It references baseline limits such as keeping activation within ±3%, bounding visitor revenue drops under 2% over 48 hours, and capping related support tickets under 5%.
Source text
SKILL.md · Apache-2.0
--- name: experiment-design-kit description: Toolkit for structuring hypotheses, variants, guardrails, and measurement plans. --- # Experiment Design Kit Skill ## When to Use - Translating raw ideas into testable hypotheses with clear success metrics. - Ensuring experiment briefs include guardrails, instrumentation, and rollout details. - Coaching pods on best practices for multi-variant or multi-surface tests. ## Framework 1. **Problem Framing** – define user problem, business impact, and north-star metric. 2. **Hypothesis Structure** – "If we do X for Y persona, we expect Z change" with assumptions. 3. **Measurement Plan** – primary metric, guardrails, min detectable effect, power calc. 4. **Variant Strategy** – control definition, variant catalog, targeting, and exclusion rules. 5. **Operational Plan** – owners, timeline, dependencies, QA/rollback steps. ## Templates - Experiment brief (context, hypothesis, design, metrics, launch checklist). - Guardrail register with thresholds + alerting rules. - Variant matrix for surfaces, messaging, and states. - **GTM Agents Growth Backlog Board** – capture idea → sizing → prioritization scoring (ICE/RICE) @puerto/README.md#183-212. - **Weekly Experiment Packet** – includes KPI guardrails, qualitative notes, and next bets for Marketing Director + Sales Director. - **Rollback Playbook** – pre-built checklist tied to lifecycle-mapping rip-cord procedures. ## Tips - Pressure-test hypotheses with counter-metrics to avoid local optima. - Document data constraints early to avoid rework during build. - Pair with `guardrail-scorecard` to ensure sign-off before launch. - Apply GTM Agents cadence: Monday backlog groom, Wednesday build review, Friday learnings sync. - Require KPI guardrails per stage (activation, engagement, monetization) before authorizing build. - If a test risks Sales velocity, include Sales Director in approval routing per GTM Agents governance. ## GTM Agents Experiment Operating Model 1. **Backlog Intake** – ideas flow from GTM pods; Growth Marketer tags theme, objective, expected impact. 2. **Prioritization** – score with RICE + qualitative "strategic fit" modifier; surface top 3 bets weekly. 3. **Design & Instrumentation** – reference Serena/Context7 to patch code + confirm documentation. 4. **Launch & Monitor** – use guardrail-scorecard to watch leading indicators (churn, complaints, latency). 5. **Learning Loop** – run Sequential Thinking retro; document hypothesis, result, decision, follow-up in backlog card. ## KPI Guardrails (GTM Agents Reference) - Activation rate change must stay within ±3% of baseline for Tier-1 segments. - Revenue per visitor cannot drop more than 2% for more than 48h. - Support tickets tied to experiment variant must remain <5% of total volume. ## Weekly Experiment Packet Outline ``` Week Ending: <Date> 1. Portfolio Snapshot – tests live, status, KPI trend (guardrail vs actual) 2. Key Wins – hypothesis, uplift, next action (ship, iterate, expand) 3. Guardrail Alerts – what tripped, mitigation taken (rollback? scope adjust?) 4. Pipeline Impact – SQLs, ARR influenced, notable customer anecdotes 5. Upcoming Launches – dependencies, owners, open questions ``` Share packet with Growth, Marketing Director, Sales Director, and RevOps to mirror GTM Agents's cross-functional communication rhythm. ---
© gtmagents · Apache-2.0 · gtmagents/gtm-agents