Product Idea Agent Skill Pack

A Claude Code plugin that replicates end-to-end product idea validation. Transform raw startup concepts into actionable validation reports with demand signals, market sizing, competitive analysis, and go-to-market strategies.

Overview

This skill pack provides 12 composable skills that guide you through a comprehensive product validation workflow:

1. Idea Brief Creation - Structure raw concepts into validated hypotheses
2. Demand Signal Research - Discover search trends and community interest
3. Problem-Segment Analysis - Define ICP and validate problem severity
4. Competitive Landscape Mapping - Identify competitors and market gaps
5. Market Sizing - Estimate TAM/SAM/SOM with growth projections
6. Pricing & WTP Research - Analyze pricing models and willingness-to-pay
7. Solution Wedge Definition - Scope MVP and differentiation strategy
8. GTM Channel Planning - Identify acquisition channels and growth loops
9. Risk Assessment - Evaluate technical, market, and execution risks
10. Scorecard Generation - Calculate multi-dimensional opportunity scores
11. Validation Report - Compile comprehensive validation deliverable
12. Orchestrator - Run the complete validation workflow end-to-end

Quick Start

Installation

The pack is a Claude Code plugin served from this repository's marketplace. In a Claude Code session:

/plugin marketplace add Jibbscript/product-idea-agent
/plugin install product-idea-agent@jibbscript

Update with claude plugin update product-idea-agent@jibbscript, remove with claude plugin uninstall product-idea-agent@jibbscript. If you installed an earlier version by copying skills/* into ~/.claude/skills/, remove those copies first; the Installation Guide shows how, along with team-wide enablement and --plugin-dir for local development.

Usage

Start a full validation by passing your idea to the orchestrator:

/product-idea-agent:idea-validation-orchestrator An AI-powered app that finds home energy leaks with a phone's thermal camera

Describing the idea in plain language works too; the orchestrator triggers on requests like:

Help me validate my startup idea for an AI-powered home energy audit app

Run or redo a single step by its namespaced command:

/product-idea-agent:competitive-landscape

Every artifact is written to the current project directory, and an interrupted validation resumes from the artifacts already there.

Skill Catalog

Skill (product-idea-agent:)	Purpose	Output Artifact
idea-brief-creator	Structure raw idea into hypothesis	idea_brief.md
demand-signals	Research search trends & community signals	signals.md
problem-segment	Define ICP & validate problem severity	icp.yaml
competitive-landscape	Map competitors & identify gaps	competitors.csv
market-sizing	Estimate TAM/SAM/SOM	market_size.md
pricing-wtp	Research pricing & willingness-to-pay	pricing.yaml
solution-wedge	Define MVP scope & differentiation	mvp_spec.md
gtm-channels	Plan acquisition channels	gtm_plan.md
risk-assessment	Identify risks & dependencies	risks.md
scorecard-generator	Calculate opportunity scores	scorecard.json
validation-report	Generate final report	validation_report.md
idea-validation-orchestrator	Run complete workflow	All artifacts

Artifact Flow

idea_brief.md
    │
    ├──► signals.md ──► market_size.md ──► gtm_plan.md
    │
    ├──► icp.yaml ──► mvp_spec.md ──────────┘
    │
    └──► competitors.csv ──► pricing.yaml
                                   │
                                   ▼
              gtm_plan.md ──► risks.md ──► scorecard.json ──► validation_report.md

Documentation

• Installation Guide - Detailed setup instructions
• Usage Guide - How to use each skill
• Contributing - How to contribute new skills
• Security - Security best practices

Evaluation

Three layers check the plugin:

# Structure: manifests, marketplace and skill frontmatter (free, every PR)
claude plugin validate . --strict
claude plugin validate .claude-plugin/plugin.json --strict

# Prompt surface: fitness benchmark and its integrity tests (free, every PR)
python3 eval/fable_prompt_bench.py
python3 -m unittest eval/test_fable_prompt_bench.py

# Behavior: plugin eval suite against a no-plugin baseline (paid, on demand)
claude plugin eval . --tag offline --scaffold --allow-tools Write WebSearch WebFetch

The eval cases live in eval/cases/ (tags smoke, offline, research, full); see eval/AGENTS.md for which flags each tag needs.

License

Apache-2.0 - See LICENSE for details.

Acknowledgments

Each skill uses the agentskills.io SKILL.md format. The pack is distributed and supported as a Claude Code plugin, which is what lets every skill resolve its artifact contract.