Claude Code Discover

A Claude Code plugin that turns a feature idea or product question into an evidence-backed product requirements document (PRD) before implementation begins. It helps you decide what to build, test the assumptions that could change that decision, and carry the evidence into implementation.

Discovery work stays in your repository so later sessions can pick up the evidence, decisions, and open risks behind the current direction. Each recipe stands alone, so you can start with the question or artifact you already have and run only the part you need.

What You Can Do

• turn user research, product ideas, or an existing codebase into Opportunities and testable hypotheses
• validate value, usability, feasibility, and viability with a method suited to the risk
• generate a self-contained HTML prototype when a usability question needs one
• define shared product structure when prototypes or downstream design need it
• produce a reviewed PRD with explicit scope, exclusions, remaining assumptions, and traceable evidence

Works standalone, or paired with claude-code-workflows for a full discovery-to-implementation cycle:

[claude-code-discover]  →  Reviewed PRD + validation evidence  →  [claude-code-workflows]
   Discovery                                                     Implementation

Quick Start

Requires Claude Code with plugin marketplace and Skills support.

# Start Claude Code
claude

# Add the marketplace and install the plugin
/plugin marketplace add shinpr/claude-code-discover
/plugin install discover@claude-code-discover
/reload-plugins

Then start with a plain request:

Goal	Command
Explore a product problem	/discover:recipe-discover Explore why trial users abandon onboarding
Validate an existing hypothesis	/discover:recipe-validate HYPO-001
Create a PRD	/discover:recipe-define Create a PRD from the validated onboarding hypotheses

For a new product or strategic reset, start with /discover:recipe-vision instead. You can also begin from an existing Opportunity or hypothesis.

Choose Where to Start

Idea or existing evidence
          ↓
Discover → Opportunities + hypotheses
          ↓
     Validate by risk ↔ Blueprint when shared structure is needed
          ↓
 Reflect completed evidence or Define a reviewed PRD

Recipe	Use it when	Result
/discover:recipe-vision	You are starting a product or revisiting its direction	Vision, outcomes, and product principles
/discover:recipe-persona	You need a grounded view of who uses the product and in what context	Evidence-backed personas
/discover:recipe-discover	You have a problem, idea, research, or an existing product to examine	Opportunities, hypotheses, and validation priorities
/discover:recipe-validate	A hypothesis could change readiness or scope	Validation evidence and an updated hypothesis
/discover:recipe-blueprint	Prototypes or downstream design need shared product structure	IA, user flows, content model, and brand direction
/discover:recipe-refine-visuals	A prototype or downstream design workflow needs reproducible visual values	Concrete tokens in brand-direction.md
/discover:recipe-reflect	Validation results or product evidence should become reusable learning	Updated evidence, learnings, and discovery index
/discover:recipe-define	You are ready to make an implementation decision	An independently reviewed PRD

How Validation Works

Validation starts by identifying the assumption most likely to change the current product decision, then selects the smallest useful test for that uncertainty.

Risk	Typical validation
Value	Market research, interviews, competitive analysis, or a demand test
Usability	A self-contained HTML prototype and user testing
Feasibility	Repository analysis, dependency review, or an isolated code spike
Viability	Business-model, financial, or regulatory analysis

Every validation updates the hypothesis with its result and evidence. A prototype is produced only when the selected usability test needs one.

Prototype Generation

For usability hypotheses, recipe-validate can generate a single self-contained HTML prototype in docs/discovery/prototypes/. It reads the hypothesis and the relevant product, persona, and design context already in the repository, then builds the interaction needed for the test.

The prototype includes its CSS, JavaScript, and mock data in one file. Open it directly in a browser without installing dependencies or running a build.

Connecting to Implementation

recipe-define compares the proposed work's cost, risk, and reversibility with the available evidence. It asks for more validation only when the result could change implementation readiness or scope. It then writes the PRD to docs/prd/ and runs an independent review before asking for approval. Prototypes from usability validation can accompany the PRD as design references.

The PRD can be handed to claude-code-workflows for design and implementation:

/dev-workflows:recipe-implement "docs/prd/feature-prd.md"

See the linked project for current installation instructions and workflow options.

What Stays in the Repository

Recipes create or update the relevant files under docs/. Validation results stay with their hypotheses, and PRDs retain links to the Opportunities, evidence, product decisions, and prototypes behind them.

Later sessions can use that record without reconstructing why a decision was made:

docs/
├── product/             # Vision, personas, design principles, learnings
│   └── design/          # Blueprint: IA, user flows, content model, brand direction
├── discovery/           # Opportunities, hypotheses, prototypes, journeys
│   └── INDEX.md         # Auto-maintained summary of discovery status
└── prd/                 # PRDs ready for implementation

Agents

Five agents run in separate contexts for independent analysis:

Agent	What it does	Why it runs in a separate context
prd-reviewer	Checks each PRD against its outcome, evidence, scope, and downstream needs	Re-checks evidence and boundaries outside the authoring conversation
codebase-analyzer	Maps decision-relevant features, user roles, and architecture from code	Separates what the product currently does from evidence about what users need
hypothesis-verifier	Decomposes hypotheses into assumptions and designs falsifiable tests	Challenges the hypothesis outside the authoring conversation before validation begins
knowledge-distiller	Extracts patterns across multiple hypothesis results	Finds cross-cutting learnings without being anchored to any single hypothesis
prototype-generator	Generates HTML prototypes from design context files	Builds product UIs isolated from test design details

License

MIT