evidence-rubric

name	evidence-rubric
description	Score a product idea against the 100-point evidence rubric before any PRD work. Eight axes: ICP specificity, recent painful event, current workaround, repetition, economic pain, switching trigger, MVP narrowness, and acquisition path to first 5 users. Returns build/interview/pivot/hold decision plus the specific axes that are weak. Use when a founder or PM is excited about an idea but evidence is thin, or before approving any spec-driven coding workflow (Spec-Kit, Kiro, GStack, Superpowers).
argument-hint	[idea to score]
allowed-tools	["Read","Write","Bash"]
model	inherit
hooks	{"Stop":[{"type":"command","command":"python3 hplan/scripts/generate_report.py harness/evidence/last_input.json --json 2>/dev/null | tail -50 || true"}]}

Evidence Rubric — 100-Point Idea Scoring

Running for: $ARGUMENTS

Core Goal

• 아이디어를 100점 루브릭으로 측정하여 "PRD 쓰기 전에 인터뷰가 더 필요한지" 명확한 신호를 만든다.
• 8개 축 각각의 점수 + 부족한 axis 목록을 반환해서, 다음 인터뷰에서 어떤 신호를 들어야 하는지 좁힌다.
• LLM의 직감 평가가 아닌 결정론적 스크립트(generate_report.py)로 점수화해 hand-wave를 차단한다.

Trigger Gate

Use This Skill When

• 사용자가 아이디어 한 문장 + 타깃 + 가설 + 대체재 + 기능 후보를 제시했을 때
• Spec-Kit / Kiro / GStack / Superpowers 워크플로우 진입 전
• 이미 인터뷰 노트가 있어 evidence strength를 객관적으로 측정하고 싶을 때
• "build로 가야 할까 interview를 더 해야 할까" 고민이 등장했을 때

Route to Other Skills When

• 점수가 낮고 인터뷰 자체가 부족할 때 → interview-synthesis (hplan plugin)
• 이미 점유된 영역으로 보일 때 → exclusions (hplan plugin) check
• 점수는 충분한데 비용 구조가 불확실할 때 → cogs-sentinel (hplan plugin)
• 아이디어 발굴 단계로 돌아가야 할 때 → opp-tree (discover plugin)

Boundary Checks

• ❌ 이 skill은 아이디어 발굴이 아니다 (그건 discover/opp-tree). 이미 아이디어가 있을 때만 호출.
• ❌ 이 skill은 PRD 작성을 허락하지 않는다. 점수가 충분해도 Product Gate + Build Gate를 거쳐야 한다.
• ❌ 점수만 높고 인터뷰가 0건이면 interview 결정이 강제된다.

Inputs

JSON 파일 또는 인라인 입력:

{
  "idea": "한 문장 가설",
  "target": "ICP 행동 기술 (인구통계 금지)",
  "hypothesis": "현재 상황",
  "alternatives": "대체재 콤마 구분",
  "features": "MVP 기능 후보 콤마 구분",
  "interview_notes": "인터뷰 발화 한 줄당 하나"
}

Steps

1. Read examples/good-01.md to internalize the rubric.
2. If user input is freeform, structure it into the 6 fields above.
3. Save to harness/evidence/last_input.json.
4. Run python3 hplan/scripts/generate_report.py <path> --json.
5. Report score + decision + breakdown + missing axes.
6. If decision == "interview", immediately route to interview-synthesis skill.
7. If decision == "build", write the report to harness/evidence/report.md and route to cogs-sentinel.

Outputs

• harness/evidence/report.md — markdown diagnosis
• harness/evidence/last_input.json — preserved input
• Decision: build (≥75 + interview_lines ≥ 2 + economic pain) / interview (≥55, or ≥75 without required conditions) / pivot (35–54) / hold (<35) — build requires mandatory economic_pain + 2+ interview lines; score alone is not sufficient

Verification

• Score is a number 0–100.
• missing array lists axes scoring < 55% of max.
• If decision == build, interview_notes has ≥ 2 lines AND combined contains economic-pain keywords.

Rubric (locked, 100 pts)

Axis	Max	Why
ICP specificity (behavior, not demographics)	20	Generic personas survive any pivot — they are noise
Recent painful event (last 30 days)	15	"I would use" → noise. "It happened yesterday" → signal
Current workaround	15	The cheapest sign of real demand: people pay time/money already
Repetition / frequency	10	One-time problems don't pay
Economic pain (money/risk/opportunity)	15	Time alone is too cheap to monetize
Switching trigger	10	"What would make you stop using your current tool?"
MVP narrowness	10	More than 3 features = no MVP
Acquisition path to first 5 users	5	If you can't name the next 5, build delay