launch-experiment — Claude Code command: ship a hypothesis as a test
launch-experiment turns an approved hypothesis into an instrumented experiment with guardrails and a rollout schedule. Use it when the design is signed off and the test needs to go live.
Apache-2.0 ★ 450 updated 2026-04-03
.claude/commands/launch-experiment.mdWhat launch-experiment does
The command takes a hypothesis ID and assembles a full experiment around it: readiness check, variant build, rollout configuration, QA, launch. You pass the surface (onboarding, pricing page, lifecycle email, in-app), variant count, and a ramp schedule like 5,25,50,100.
Workflow has five stages. It confirms design sign-off, instrumentation coverage and guardrails first. Then it pulls specs, assets and targeting rules for each arm, configures the experimentation platform with ramp and alerts, runs smoke tests with screenshots, and activates the test with telemetry dashboards and channel notifications.
It depends on agents and skills: test-engineer builds variants and QA evidence, experimentation-strategist confirms governance, guardrail-scorecard validates thresholds, experiment-design-kit enforces templates. Missing pieces break those checks.
Invocation: /growth-experiments:launch-experiment --id EXP-142 --surface onboarding --variant-count 3 --ramp 5,25,50,100.
What it helps with
- assemble test variants from specs and targeting rules
- set up a ramp schedule with alerts on the experimentation platform
- run smoke tests and capture screenshots for approval
- produce a launch packet with QA evidence and timeline
- prepare the stakeholder announcement and escalation matrix
Example prompts for the agent
- /growth-experiments:launch-experiment --id EXP-142 --surface onboarding --variant-count 3 --ramp 5,25,50,100
- Launch experiment EXP-142 on onboarding with three variants and ramp 5,25,50,100
- Run the readiness check for EXP-142 and build the launch packet
How to install launch-experiment
Manually: put the files in place
.claude/commands/launch-experiment.mdCall it with /launch-experiment. For all projects use ~/.claude/commands/
Manually: put the files in place
~/.codex/prompts/launch-experiment.mdCall it with /prompts:launch-experiment. Codex marks this format deprecated in favour of skills
Manually: put the files in place
.github/prompts/launch-experiment.prompt.mdManually: put the files in place
.gemini/commands/launch-experiment.tomlGemini stores commands as TOML: the prompt text goes into the prompt field
Questions about launch-experiment
How do I install launch-experiment in Claude Code?
Create .claude/commands/launch-experiment.md in your project and put the command content there. It then becomes available as /growth-experiments:launch-experiment.
What does launch-experiment do?
It converts an approved hypothesis into a test: checks readiness, assembles variants, configures the rollout, and launches with monitoring. Outputs include a launch packet, platform configuration, and a stakeholder announcement.
Which agents and skills does it need?
It calls test-engineer for variant assembly and QA, experimentation-strategist for governance, plus the guardrail-scorecard and experiment-design-kit skills. Without them, threshold validation and best-practice checks will not run.
Source text
launch-experiment.md · Apache-2.0
--- name: launch-experiment description: Converts an approved hypothesis into a fully-instrumented test with guardrails and rollout plan. usage: /growth-experiments:launch-experiment --id EXP-142 --surface onboarding --variant-count 3 --ramp 5,25,50,100 --- # Command: launch-experiment ## Inputs - **id** – experiment or hypothesis identifier. - **surface** – product area/channel (onboarding, pricing page, lifecycle email, in-app). - **variant-count** – number of variants/arms including control. - **ramp** – comma-separated rollout schedule (%) or JSON file reference. - **holdout** – optional holdout/ghost-experiment definition for measurement. - **notes** – free-text for special approvals or exception handling. ## Workflow 1. **Readiness Check** – confirm design sign-off, instrumentation coverage, and guardrails. 2. **Variant Assembly** – pull specs, assets, and targeting rules for each arm. 3. **Rollout Plan** – configure flag/experimentation platform with ramp schedule + alerts. 4. **QA & Approvals** – run smoke tests, capture screenshots, and gather stakeholder approval. 5. **Launch & Monitoring** – activate test, enable telemetry dashboards, and notify channels. ## Outputs - Launch packet with specs, QA evidence, approvals, and rollout timeline. - Experiment platform configuration export + guardrail monitors. - Stakeholder announcement + escalation matrix. ## Agent/Skill Invocations - `test-engineer` – builds variants, instrumentation, and QA evidence. - `experimentation-strategist` – confirms governance + approvals. - `guardrail-scorecard` skill – validates guardrail coverage + thresholds. - `experiment-design-kit` skill – ensures templates + best practices are applied. ---
© gtmagents · Apache-2.0 · gtmagents/gtm-agents