market-validate

End-to-end market validation. Produces ./growth/01-research/.

When to invoke

• User asks: "validate this product", "is this idea worth building", "run a market analysis", "should I keep building this".
• Invoked by the /growth orchestrator as phase 1 of 3.
• Standalone via /growth-market.

Inputs the skill auto-detects

Before asking the user anything, scan:

1. Repo signals:
   • README.md (root + any other) — product description, tagline, audience hints.
   • package.json — description, homepage, keywords, dependencies (reveal stack / category).
   • .planning/ — if present, scan PROJECT.md, ROADMAP.md, current milestone files for explicit positioning, ICP, pricing.
   • /docs, /marketing, /content — any in-repo marketing copy.
   • src/app/page.* / src/pages/index.* / index.html — extract hero copy, value prop.
   • Any pricing.md, pricing.json, tiers.* files.
   • Locale files (messages/*.json, locales/*, i18n/*) — reveal target markets.
2. Live URL (if findable):
   • package.json homepage field.
   • Linked URLs in README.
   • .env.example / .env.production — NEXT_PUBLIC_SITE_URL, VITE_SITE_URL, etc.
   • If found, WebFetch the live URL to extract: hero, features, pricing, social proof, blog topics if any.
3. Optional user brief — if the orchestrator passed a --brief "..." arg, use it.

Adaptive Socratic questioning

After auto-scan, the skill identifies gaps and asks 5–10 questions max via AskUserQuestion. Focus on what is NOT inferable from code/site:

• ICP confirmation (you saw segment X — confirm or refine).
• Pricing model + tiers (if not visible).
• Top 3 competitors the user actively watches.
• North-star metric the user cares about (signups / revenue / retention / impressions).
• Founder/team context (solo, side project, funded — affects financial framework).
• Any market/regulatory constraints (e.g. education vertical, fintech).
• Anything explicitly OFF the table (e.g. "I refuse to compete on price").

Use detected codebase + live-URL signals to anchor questions. Never ask things the skill could have inferred.

Under --auto mode

Skip this entire questioning phase. For each question that would have been asked, infer the most plausible answer from auto-detected signals and record the inferred value + source in a top-level ## Auto-mode assumptions section at the END of research.md.

Example log entries:

• "ICP inferred: Israeli high-school CS students preparing for 5-unit Bagrut. Source: trilingual locale files (he/ar/en), .planning/PROJECT.md line 14, hero copy."
• "Pricing inferred: two-tier ~99₪ basic + premium. Source: .planning/PROJECT.md, no live pricing page."
• "Competitors inferred: <Comp-A>, <Comp-B>, <Comp-C>. Source: WebSearch for category + Reddit threads. Mark as low-confidence."
• "North-star metric inferred: paying users (from explicit 2500-user target in plugin invocation). Source: orchestrator args."
• "Founder context inferred: solo builder. Source: git log author uniformity + .planning notes."

Confidence-tag every inferred value as high / medium / low. Anything low should appear in verdict.json top_risks[] so the user sees what's uncertain.

Framework battery (full, every run)

Run all 8 frameworks from ${CLAUDE_PLUGIN_ROOT}/references/frameworks/:

1. 01-market-validation.md — Lean Startup hypothesis, Sean Ellis PMF test (40% must-have), Validation Board.
2. 02-business-models.md — Business Model Canvas (9 blocks), Porter's 5 Forces, SWOT.
3. 03-financial-analysis.md — Unit economics, CAC/LTV plausibility, runway, pricing sensitivity.
4. 04-competitive-analysis.md — Direct competitors, indirect competitors, alternatives, defensive moats.
5. 05-startup-metrics.md — PMF signals, retention curves, viral coefficient potential, SaaS metrics if applicable.
6. 06-data-sources.md — Which real-world data sources support / refute the thesis.
7. 07-ai-tools.md — AI landscape context: what AI advantages does the product have, what AI threats.
8. 08-funding-planning.md — Bootstrappable vs needs funding, valuation comparables, signal-to-investors.

For each framework: read the reference file, apply to the product, produce a 3-5 paragraph section in research.md.

External research (free sources only)

During analysis:

• WebSearch: top 3-5 competitors by category, recent industry reports, Reddit/HN threads on the problem space.
• WebFetch: competitor homepages, pricing pages, G2/Capterra/Trustpilot review pages (if accessible), top Reddit threads found.

Surface findings directly in research.md with citations (link inline).

Outputs

./growth/01-research/research.md

Structure:

# Market Validation: <Product Name>

## Executive Summary
(One paragraph: what is the product, who it's for, the verdict, the score, the headline reason.)

## Product Snapshot
- Name, category, primary market, URL if any.
- Inferred ICP (segment + persona detail).
- Inferred positioning + value prop.

## Framework Analysis
### 1. Market Validation
### 2. Business Model
### 3. Financial Analysis
### 4. Competitive Analysis
### 5. Startup Metrics
### 6. Data Sources
### 7. AI Landscape
### 8. Funding Strategy

## Competitor Teardown
Top 3-5 with: positioning, pricing, strengths, weaknesses, what they do that you should learn from.

## PMF Signals
Concrete leading indicators to watch.

## Top Risks (3-5)
With severity + mitigation each.

## Top Opportunities (3-5)
With leverage + concrete action each.

## Verdict
PROCEED / PIVOT / KILL — with 2-4 sentence rationale.

## Score: X.X / 10
With 1-sentence per dimension breakdown.

## Sources
Numbered list of all external URLs cited.

./growth/01-research/research.html

Only generated if --md-only flag is NOT set. Rendered styled to codebase design tokens.

Codebase style detection:

1. Look for tailwind.config.{js,ts} → extract colors + fonts.
2. Look for :root { --color-* } CSS variables in any *.css under src/ or styles/.
3. Look for theme.{js,ts} in styled-components or emotion setups.
4. Fall back to default editorial dark (deep indigo #0f1226 bg, teal #1e88a8 accent, soft cream text).

HTML structure: single-page with nav anchored to sections, expandable framework cards, verdict callout box at top, sources at bottom.

./growth/01-research/verdict.json

Conform to ${CLAUDE_PLUGIN_ROOT}/references/output-schemas.md §1.

Sidecar writes to shared references/

Write these so content-month and seo-engine inherit them:

• ./growth/references/icp.md — synthesized ICP: segment, demographics, psychographics, jobs-to-be-done, top 3 pains, top 3 desired gains, where they hang out online.
• ./growth/references/positioning.md — positioning statement: For [ICP] who [need], [product] is the [category] that [unique value], unlike [alternatives], we [differentiator].
• ./growth/references/competitors.md — top 3-5 competitors with positioning + pricing + URLs (used by seo-engine for /vs and /alternatives pages).

Scoring rubric (the score)

Score 1-10 = average of 5 dimensions, each 1-10:

Verdict thresholds:

• 8.0+: PROCEED (high confidence)
• 6.0–7.9: PROCEED with notes (specify the conditions)
• 4.0–5.9: PIVOT (which dimension is broken, what direction)
• < 4.0: KILL (or major reframe)

Flags

• --auto — unattended mode. Skip Socratic Q&A entirely; infer all values from auto-detected signals; log assumptions in research.md § Auto-mode assumptions; low-confidence inferences appear as top_risks[] in verdict.json.
• --md-only — suppress research.html generation.
• --brief "..." — user-supplied product description (skips some auto-detection).
• --url <url> — override live URL detection.

Verification

After running, the skill should self-check:

• research.md ends with a verdict and a numeric score.
• verdict.json is valid against the schema in output-schemas.md §1.
• research.html exists (unless --md-only).
• references/icp.md, positioning.md, competitors.md are all written.
• All external URLs cited in ## Sources resolved successfully during research.