User Story

As a solo founder running ApexYard, I want a lightweight 5-question idea-validation skill I can invoke either standalone (/validate-idea IDEA-NNN) or as an offered follow-up step inside /idea and /handover, so that I have a 10-minute pre-spec gate that catches "this idea isn't worth a PRD" without needing event storming, Wardley mapping, or any heavyweight strategy ceremony.

Summary

New skill at .claude/skills/validate-idea/SKILL.md. Five questions (target user, current alternative, smallest version, kill criteria, build/buy/rent), asked one at a time, ending in a GREEN / YELLOW / RED verdict. RED automatically updates the IDEA-NNN row in projects/ideas-backlog.md to WONTDO.

Two integrations:

• /idea — at the end of the capture flow (after the optional GitHub Issue offer), a third optional step: "Validate now? Run /validate-idea IDEA-NNN — y/n (default n)". Default-no respects the lightweight-capture intent.
• /handover — at the end of the integration-plan emit, a conditional offer gated on a dormancy heuristic (last commit > 90 days AND zero open PRs AND no recent issue activity). Active projects don't see the prompt; dormant ones do.

Why this and not something heavier

The user-and-I pair-debated the "should we add validation tooling?" question and rejected event storming + Wardley mapping as too heavy for the audience (solo founders). Both are 4–5-stakeholder workshops, neither fits the "single-Claude-session, 10 minutes max" shape ApexYard's other skills hold to (/idea, /decide, /write-spec).

The lighter alternative — five plain questions with a one-page output — fits. It nudges the user to think falsifiably (Q4 — kill criteria) and surface the build-or-rent call early (Q5), without imposing a methodology.

Testing

Skills are markdown specs — no shell test fixture. Conformance is exercised by the agent following the spec.

Smoke check on the skill structure:

# Front-matter has the standard fields
head -10 .claude/skills/validate-idea/SKILL.md

# Five questions, one per turn — verify the spec calls out "ONE AT A TIME"
grep -c "ASK ONE AT A TIME\|one at a time" .claude/skills/validate-idea/SKILL.md
# Expect: ≥ 1

# Verdict path includes side effects on RED
grep -A 1 "RED.*backlog" .claude/skills/validate-idea/SKILL.md

The actual dogfood will be the next idea I capture — the spec is followable as-is.

Scope — what this does NOT do

• No automated competitive research. Q2 ("what do they do today?") is the user's own observation, not a skill that crawls the web.
• No scoring rubric. Verdict is a PM judgement call (Product Manager role activates for the read-out). Tried scoring rubrics; they're either too coarse or too detailed for one page.
• No auto-trigger of /write-spec on GREEN. The user makes that call separately. Decoupling validation from spec-authoring keeps each step's intent explicit.
• No event storming, no Wardley mapping. Out of scope by design — wrong methodology weight for solo founders. Captured in the skill's "Out of scope" + the issue body.

Glossary

Closes #130