ai-startup-skills
AI 제품을 만드는 스타트업을 위한 7단계 에이전트 스킬: 검증 → 비즈플랜 → 정책 → 로드쇼 → 고객 확보 → 가격 책정 → 등록
저장소 정보는 아직 수집하지 않았습니다.
아직 감지된 업데이트가 없습니다. GitHub 릴리스가 나오면 여기에 보입니다.
메이커가 쓴 소개는 위의 한 줄이 전부입니다. 더 자세한 내용은 저장소 README에서 읽을 수 있습니다. 저장소 README ↗
README 전체를 불러오는 중입니다.
README 전문 보기 ↗Value matching and promise-tracking trust scores for every race on your ballot. hushpolitics.com
Next.js 16 (App Router) · Supabase · Vercel
npm install
cp .env.example .env.local # optional — the app runs without it
npm run dev
With no env vars the app serves the seed dataset in src/lib/seed-data.ts.
That is intentional: the first deploy works before the backend exists, and the
same fixtures are what the ingestion pipeline is tested against.
src/
app/ routes — one per view in the design
page.tsx Feed
compare/ Vote compare + stance grid
your-ballot/ Countdown, races, polling place
hush-guide/ Address + issues -> race tile grid, comparisons, bills
fact-check/ Claims and verdicts
profile/ The "you" surface (topics, ranked issues, saved)
politician/[id]/ Profile
politician/[id]/trust Promise ledger
login/ pricing/ Auth and the (not yet live) membership page
api/ingest/ Cron-driven ingestion worker
auth/callback/ Magic-link exchange
components/
AppShell.tsx Sidebar, top bar, footer
ui.tsx Design-system primitives
Gate.tsx Subscription gate (off until pricing is decided)
views/ One component per screen
lib/
theme.ts Design tokens, ported from the Claude Design canvas
types.ts Domain types
seed-data.ts Generated fixtures — do not hand-edit
repo.ts Data access seam: swap to Supabase here, views unchanged
scoring.ts Trust and value-match maths (mirrors the SQL)
prefs.tsx Per-user state (ranked issues, saved, ZIP)
supabase/ Browser, server and service-role clients
pipeline/ Source rubric + claim assessment
supabase/migrations/ 0001 schema · 0002 RLS · 0003 ingestion+scoring · 0004 billing
docs/ingestion-pipeline.md How the AI data layer works and why
Views never import seed-data directly — they go through lib/repo.ts. When
the Supabase tables are populated, each function in repo.ts swaps its body for
a query and nothing in components/ changes.
Supabase dashboard → SQL Editor → run supabase/migrations/*.sql in order.
Or with the CLI:
npx supabase link --project-ref <ref>
npx supabase db push
60% promises kept, 25% recency, 15% promise significance. The weights live in
the scoring_weights table rather than in code, so a rubric change is a dated
row and the number on screen can always be explained. See
compute_trust_score() in migration 0003.
The map on Your Ballot is a live Google Maps embed centered on the user's
address (whatever they've most recently given the app — the polling-place
lookup, their onboarding street address, or the city/state/ZIP set via the
header pill). It needs NEXT_PUBLIC_GOOGLE_MAPS_API_KEY (Maps Embed API
enabled) in the environment — see .env.example. Without it, that panel
falls back to an "Open in Google Maps" link rather than a broken embed, so
missing setup is visible instead of silently failing.
vercel.json runs /api/ingest once daily at 06:00 UTC (2am ET). That cadence
is a plan constraint, not a design choice: Vercel Hobby allows one cron run
per day, and anything more frequent fails the build outright with
would run more than once per day. On Pro, raise it — every 4 hours
(0 */4 * * *) is the intended cadence once there are real sources to crawl.
Note also that vercel.json rejects unknown keys inside a cron entry, so the
schedule cannot carry an inline comment. Hence this section.
The model never writes a published row. It writes to claim_assessments with
its evidence and confidence; shouldAutoPublish() in
src/lib/pipeline/assess.ts decides whether that clears the bar or goes to
review_queue. Read docs/ingestion-pipeline.md before changing any of it —
the bias and defamation notes in there are load-bearing, not boilerplate.