From vibe-innovation
Designs single validation experiments testing one falsifiable assumption with a numeric threshold and method from a catalog. Use between phases, for Phase 3 refinements, or Phase 4 follow-ups.
How this skill is triggered — by the user, by Claude, or both
Slash command
/vibe-innovation:innovate-experimentThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
**Assumption to test:** $ARGUMENTS
Assumption to test: $ARGUMENTS
Read the following files in order before acting.
${CLAUDE_SKILL_DIR}/../../docs/validation_methods.md (catalog of pretotypes for technical artifacts and analogous methods for institutional artifacts)${CLAUDE_SKILL_DIR}/../../docs/phase_3_value_architecture.md Step 5 (the canonical experiment design pattern)${CLAUDE_SKILL_DIR}/../../docs/principles_and_antipatterns.md (Principle 7: one experiment, one assumption)If the argument is an assumption ID (for example, A3), read the current ICD Section 3.3 (Assumption map) and load the assumption. If the argument is free text, draft an assumption statement together with the user and ask whether to register it in Section 3.3 first. If the argument is empty, list the top 3 untested assumptions in Section 3.3 by priority score and ask which one to design for.
The assumption must be a single, falsifiable claim. If it bundles two claims (for example, "users want this AND will pay for it"), split it into two assumptions and pick the more critical one.
Ask one question: what observation would convince the team this assumption is false? Until the team can name a falsifier, do not proceed. An experiment without a falsifier is theatre.
From validation_methods.md, pick the cheapest method that can produce the falsifier. The catalog covers, at minimum:
If none of these fit, name the rationale and pick a custom method.
The metric must be numeric and observable from the outside. "People liked it" does not count. "At least 30 signups in 7 days from a population of 500 visitors, click-through rate above 5 percent" counts.
The metric must include both a numerator and a denominator where applicable. A bare count without a sample size is uninterpretable.
Set the numeric threshold before running the experiment. State the threshold and the falsification rule in one line: "If [metric] is below [threshold], the assumption is falsified."
If the team is reluctant to commit to a threshold, this is a signal that the team is afraid of the answer. Hold the line. Pre-commit or do not run.
Three numbers: estimated effort (person-hours), estimated calendar time, and estimated direct cost in EUR. Pretotypes that take more than two weeks or more than a few hundred euros are usually disguised products. Recommend a smaller experiment.
State explicitly what happens if the experiment falsifies the assumption. Two sentences. The kill condition is part of the experiment design, not an afterthought. If the team cannot live with the kill condition, they will retroactively shift the threshold. Resolve the discomfort now.
Produce a single experiment record in the schema-compatible format:
EXPERIMENT DESIGN
=================
ID: E[N] (next available)
Assumption tested: A[N] (or new assumption text)
Pretotype type: [method from validation_methods.md]
Success metric: [numeric, with numerator and denominator]
Threshold: [numeric pre-committed value, with falsification rule]
Estimated cost and time: [hours, days, EUR]
Kill condition: [what happens if falsified]
Owner: [name]
Earliest start: [date]
Append to ICD Section 4.5 (Experiment design). If Section 4.5 does not exist yet (running pre-Phase-3), buffer the design and recommend running /innovate-phase 3 to integrate it formally.
Respond in the language the user is writing in.
npx claudepluginhub raphael310/vibe-innovation --plugin vibe-innovationProvides behavioral guidelines to reduce common LLM coding mistakes, focusing on simplicity, surgical changes, assumption surfacing, and verifiable success criteria.
Searches, retrieves, and installs Agent Skills from prompts.chat registry using MCP tools like search_skills and get_skill. Activates for finding skills, browsing catalogs, or extending Claude.
Creates, edits, and optimizes skills for Claude Code, including drafting, evaluating with test prompts, iterating on performance, and improving skill descriptions for better triggering accuracy.