Skill cro-optimization
✓ VerificadoEjecuta la optimización de la tasa de conversión mediante un proceso estructurado de auditoría, generación de hipótesis, diseño de experimentos y análisis estadístico.
- Estrellas en GitHub
- 1066
- Archivos
- 3
- Último commit
- 8 nov 2026
- Licencia
- MIT
- Herramientas permitidas
- —
¿Qué hace la skill cro-optimization?
Ejecuta la optimización de la tasa de conversión mediante un proceso estructurado de auditoría, generación de hipótesis, diseño de experimentos y análisis estadístico.
Resumen generado a partir del SKILL.md de la skill.
Sobre la skill cro-optimization
Esta SKILL permite a los equipos de growth y desarrolladores optimizar las tasas de conversión mediante una disciplina estructurada y basada en hipótesis. Guía a los usuarios a través de un ciclo completo de cuatro fases que consiste en auditoría, generación de hipótesis, diseño de experimentos y toma de decisiones final. Un agente debe cargar esta SKILL cuando un usuario necesite solucionar una caída alta en un embudo de conversión, diseñar un test A/B o interpretar resultados estadísticos ambiguos.
El proceso comienza con auditorías cuantitativas, cualitativas y heurísticas para identificar puntos de fricción. Una vez identificados, la SKILL ayuda a formular enunciados comprobables utilizando una estructura específica y los prioriza mediante puntuación ICE o PIE. Genera planes de prueba detallados en markdown que incluyen requisitos de tamaño de muestra, métricas principales y de control (guardrail metrics), y criterios de decisión. Esta SKILL está diseñada para páginas existentes con tráfico suficiente, requiriendo normalmente al menos 5,000 conversiones mensuales por variante para garantizar la significancia estadística. Es agnóstica en cuanto a herramientas, lo que significa que funciona junto a cualquier infraestructura de test A/B o configuración de analítica existente.
Lo que un agente puede hacer con ella
- 1Diagnosticar puntos de abandono en el embudoRealiza una auditoría exhaustiva de un flujo de usuario específico para encontrar por qué los clientes abandonan sus carritos.
- 2Generar hipótesis de prueba priorizadasCrea una lista de cambios comprobables basados en los hallazgos de la auditoría y clasifícalos por impacto y facilidad.
Escrito a partir del SKILL.md de la skill. Cada caso de uso se verifica con una cita del mismo.
Cómo instalar la skill cro-optimization
Cada comando copia la carpeta de la skill al directorio de skills del cliente. Reinicia el cliente o inicia una nueva sesión para cargarla.Instalar la skill cro-optimization en Plugin
/plugin marketplace add rampstackco/claude-skills /plugin install rampstack-skills@rampstack
Instalar la skill cro-optimization en Claude Code
~/.claude/skills/cro-optimization/SKILL.md
$ npx degit rampstackco/claude-skills/skills/cro-optimization ~/.claude/skills/cro-optimizationInstalar la skill cro-optimization en Claude Code (project)
.claude/skills/cro-optimization/SKILL.md
$ npx degit rampstackco/claude-skills/skills/cro-optimization .claude/skills/cro-optimizationInstalar la skill cro-optimization en Codex
~/.agents/skills/cro-optimization/SKILL.md
$ npx degit rampstackco/claude-skills/skills/cro-optimization ~/.agents/skills/cro-optimizationInstalar la skill cro-optimization en Gemini CLI
~/.gemini/skills/cro-optimization/SKILL.md
$ npx degit rampstackco/claude-skills/skills/cro-optimization ~/.gemini/skills/cro-optimizationInstalar la skill cro-optimization en Claude.ai
npx degit rampstackco/claude-skills/skills/cro-optimization cro-optimization && zip -r cro-optimization.zip cro-optimization # claude.ai → Customize → Skills → + → Create skill → Upload a skill → cro-optimization.zip
Verificaciones automáticas
Ejecutado 15 nov 2026Informe completo →Superó nuestras comprobaciones automáticas. No es una auditoría de seguridad: lee los scripts antes de dejar que un agente los ejecute. Cómo funcionan las verificaciones
- Repositorio de origen vinculado y accesiblerampstackco/claude-skills/skills/cro-optimization is public and reachablegithub.com/rampstackco/claude-skills/tree/HEAD/skills/cro-op ↗Aprobado
- Licencia de código abierto (OSI)MITspdx.org/licenses/MIT.html ↗Aprobado
- Commit o lanzamiento en el último añoLast commit 2026-11-08github.com/rampstackco/claude-skills/commits ↗Aprobado
- Sin descargar y ejecutar ni robo de credenciales en los scriptsNo scripts in the skill folderAprobado
- Sin instrucciones ocultas en SKILL.mdNo override phrases, safeguard switches, credential requests or hidden charactersgithub.com/rampstackco/claude-skills/blob/HEAD/skills/cro-op ↗Aprobado
- Las herramientas preaprobadas están acotadasNo pre-approved tools; the agent asks before using onegithub.com/rampstackco/claude-skills/blob/HEAD/skills/cro-op ↗Aprobado
- SKILL.md sigue el formato de Agent SkillsValid name and description; says when to use it; 278 linesgithub.com/rampstackco/claude-skills/blob/HEAD/skills/cro-op ↗Aprobado
SKILL.md
Ver en GitHubCRO Optimization
Run conversion rate optimization as a structured discipline: audit → hypothesize → test → decide. Stack-agnostic. Tool-agnostic.
This skill is for running tests against existing pages and flows. For writing landing page copy from scratch, use landing-page-copy. For setting up the analytics that make CRO possible, use analytics-strategy.
When to use
- Converting traffic at lower rate than expected
- Specific funnel step has high drop-off
- Pages with high traffic that could move the needle if optimized
- A/B testing infrastructure exists (or can be set up)
- Statistical significance and sample size questions
When NOT to use
- Without sufficient traffic to test (under ~5,000 monthly conversions per variant)
- Pre-launch (no users to test on yet)
- Strategy or messaging-level questions that need qualitative research first
- Brand-defining choices (CRO can't optimize a fundamentally wrong brand)
Required inputs
- The page or flow under optimization
- Current conversion rate and traffic volume
- Access to analytics (event tracking, funnel data)
- An A/B testing tool (or willingness to set one up)
- Time and budget for testing (typically 2 to 8 weeks per test)
The framework: 4 phases
1. Audit
Diagnose before treating.
Quantitative audit:
- Funnel data. Where are users dropping off? The biggest drop is the biggest opportunity.
- Segmentation. Does the funnel perform differently by source, device, geography, audience type?
- Performance data. Are slow pages dragging conversions?
- Search Console / on-site search. What are users looking for that they can't find?
Qualitative audit:
- Session replay. Watch 20+ sessions of users on the target flow. Note friction, confusion, hesitation.
- Heatmaps. Where do users click? Where do they scroll? Where do they not?
- User interviews / surveys. Why did users not convert? Survey people who started but abandoned.
- Form analytics. Which fields cause abandonment? Which cause errors?
- Customer support tickets. What conversion-related questions come in?
Heuristic audit:
- Apply CRO heuristics to the flow:
- Is the value proposition clear in 5 seconds?
- Is there a single primary CTA per page?
- Is the form length appropriate to the offer?
- Is the trust/social proof present?
- Are objections handled?
- Is the page accessible? (Accessibility issues hurt conversion silently.)
The audit produces a list of suspected friction points. Each becomes a hypothesis candidate.
2. Hypothesis
A testable statement.
Hypothesis structure:
Because [observation from audit], we believe that [change] will produce [predicted outcome] for [user segment], because [reason].
Example:
Because session replays show users abandoning at the shipping step (audit), we believe that adding visible shipping cost to the product page (change) will increase add-to-cart conversion by 5 percent (outcome) for desktop users (segment), because users are surprised by shipping cost and abandon (reason).
Hypothesis quality criteria:
- Specific change (not "improve the design")
- Measurable outcome (with a target)
- Grounded in evidence (audit, research, prior tests)
- Tied to a known mechanism (why would this work?)
Hypothesis prioritization (ICE or PIE):
- Impact: How much could this move the metric?
- Confidence: How likely is the hypothesis to be right?
- Ease: How easy to test? (Time, complexity, risk)
Score each 1 to 10. Highest combined scores test first.
3. Test design
A test that produces an unambiguous answer.
Sample size and duration:
Use a sample size calculator (most A/B tools have one) before launching. Inputs:
- Baseline conversion rate
- Minimum detectable effect (the smallest lift you'd care about)
- Statistical power (typically 80%)
- Significance level (typically 95%)
This produces required sample size per variant. Run the test until that sample is reached, OR for a minimum duration that captures full business cycle (typically 2 weeks minimum, to cover weekends and weekly patterns).
Common test setup mistakes:
- Stopping the test the moment significance is hit (peeking)
- Running tests for too short to capture a full business cycle
- Running multiple overlapping tests on the same flow
- Testing during atypical periods (Black Friday, holidays, major campaigns)
- Excluding mobile when 50%+ of traffic is mobile (or vice versa)
- Testing on too small a slice of traffic (low statistical power)
- Not segmenting analysis (overall lift can hide negative impact on a segment)
Test parameters to define before launch:
- Primary metric (one)
- Guardrail metrics (do not go down)
- Sample size
- Duration (minimum and maximum)
- Decision criteria (when to ship, when to kill, when to extend)
- Segments to analyze in addition to overall
4. Decide
After the test concludes.
Decision framework:
| Outcome | Decision |
|---|---|
| Variant clearly wins (>95% significance, exceeds minimum effect) | Ship variant. Document. Continue testing. |
| Variant clearly loses | Kill. Capture the lesson. Iterate hypothesis. |
| Inconclusive (neither significant) | Larger test, different angle, or move on. Don't ship "tied" variants. |
| Small lift, lots of variance | Probably not worth shipping. Even if "winner," may not replicate. |
| Wins overall, loses for important segment | Investigate segment. Consider segment-specific solution. |
Anti-patterns:
- "It looks like it's winning, ship it" before reaching significance
- Shipping a variant because the team wants to (HiPPO - highest paid person's opinion)
- Killing tests too early because they look bad
- Re-running tests until they "win" (false positive risk)
- Not capturing the learning when a test loses
Statistical foundations
Significance and confidence
A 95% significance level means: if there were truly no difference between variants, there's only a 5% chance you'd see results this extreme by chance.
That's not the same as "95% chance the variant wins."
Most CRO tools report Bayesian probabilities ("95% chance of being best"). Read the methodology your tool uses.
Sample size
Conversion testing needs more sample than people intuit. Quick reference:
| Baseline rate | Minimum detectable effect | Sample per variant |
|---|---|---|
| 2% | 10% relative lift | ~75,000 |
| 2% | 20% relative lift | ~19,000 |
| 5% | 10% relative lift | ~30,000 |
| 5% | 20% relative lift | ~7,500 |
| 10% | 10% relative lift | ~14,000 |
| 10% | 20% relative lift | ~3,500 |
(Approximate. Use a calculator.)
If your monthly conversions per variant don't reach these numbers, A/B testing won't produce reliable results. Iterate via design and qualitative research instead.
Multiple testing
The more variants and metrics tested simultaneously, the more false positives. Adjust significance thresholds for multiple comparisons (Bonferroni or similar).
Workflow
- Audit. Quantitative + qualitative + heuristic.
- Generate hypotheses. From audit findings. Apply hypothesis structure.
- Prioritize. ICE or PIE. Top 3 to 5 to test next.
- Design the test. Sample size, duration, primary and guardrail metrics, decision criteria.
- Implement. Build variants. QA carefully (broken variants invalidate tests).
- Run. Don't peek. Don't stop early.
- Analyze. Overall and by segment. Note interesting patterns regardless of significance.
- Decide. Ship, kill, or extend.
- Document. Hypothesis, design, results, decision, lesson.
- Compound. Apply lessons to next round of hypotheses.
Failure patterns
- Testing without audit. Random changes, random results.
- Vague hypotheses. "Make it better" is not a hypothesis.
- Peeking and early stopping. Bias toward false positives.
- Underpowered tests. Not enough sample for a real conclusion.
- HiPPO override. Highest paid person's opinion overrides the data.
- Testing during atypical periods. Holidays distort results.
- Single metric obsession. Conversion ups but average order value craters. Net loss.
- No guardrail metrics. Testing for one outcome, missing damage to others.
- Documentation gap. Wins captured, losses forgotten. Same hypothesis re-tested 3 times.
- Treating each test in isolation. Compounding learning across tests is where CRO programs really win.
Output format
Default output: a markdown test plan at cro-test-[hypothesis-slug].md per test. After the test runs, append the results section.
Structure:
# Test: [Hypothesis short name]
## Hypothesis
Because [observation], we believe that [change] will produce [outcome] for [segment], because [reason].
## Audit evidence
[What evidence supports this hypothesis]
## Test design
- Primary metric:
- Guardrail metrics:
- Sample size required:
- Duration: minimum X, maximum Y
- Variant traffic split:
- Segments to analyze:
## Decision criteria
- Ship if: [conditions]
- Kill if: [conditions]
- Extend if: [conditions]
## Results (filled after test)
- Sample reached:
- Duration actual:
- Primary metric: [variant vs control + significance]
- Guardrail metrics: [results]
- Segment analysis: [findings]
## Decision
[Ship / Kill / Extend / Iterate] - [Why]
## Lesson
[What this teaches us, regardless of outcome]
If required data is unavailable
This skill's output depends on data, measurements, or tool results it cannot generate on its own. When a required input, tool, or data source is unavailable or unverifiable, the sanctioned output is the deliverable with the gap stated: what was needed, what was actually obtained or verified, and which parts of the output are affected. Fabricating, estimating, or interpolating a required number to complete the deliverable is never sanctioned. A stated gap is a complete answer.
Reference files
references/hypothesis-library.md- Common high-impact hypothesis patterns by funnel stage.
Archivos de esta skill
3Preguntas sobre la skill cro-optimization
¿Cómo instalo la skill cro-optimization en Claude Code?
Ejecuta npx degit rampstackco/claude-skills/skills/cro-optimization ~/.claude/skills/cro-optimization. Claude Code carga las skills desde ~/.claude/skills (todos los proyectos) o .claude/skills (un proyecto) y usa una skill cuando su descripción coincide con la tarea.
¿Qué clientes pueden usar la skill cro-optimization?
Cualquier cliente que lea el formato Agent Skills: Claude Code, Codex, Gemini CLI y claude.ai (subir como ZIP). Los scripts de la skill se ejecutan solo donde el cliente permite la ejecución de código.
¿Cómo debe estructurarse una hipótesis?
Una hipótesis debe seguir un formato específico: Debido a [observación], creemos que [cambio] producirá [resultado previsto] para [segmento de usuarios], debido a [razón].