-
Both are pluggable seams (dev-plan-4.md §1.2 items 1-2), same shape as T20's sendVerificationEmail: a fixed env var gates a real vendor call that isn't implemented yet (none has been picked), and the no-vendor path is what actually runs today. describeGarmentImage() — mode 2 has no admin around to hand-write garments.appearanceDescriptor the way admin_seed garments do, so it needs *something* non-empty to fill that field; today that's a fixed placeholder string, swapped for a real vision-model call once a vendor's picked. moderateGarmentImage() — the requirements doc flags image-content moderation as vendor-TBD but never actually lists it as its own task; this fills that gap. Every call writes a moderation_log row (new dressup-store.js export, M.9 schema) regardless of which branch runs, so the audit trail exists before a real vendor does. Default-permissive when unconfigured — T2's upload endpoint already enforces the real trust-boundary checks (magic-byte type check, 8MB cap, EXIF-stripped re-encode); this only covers the *content* check, which nothing here can perform without a vendor. Neither is unit-tested standalone (server.js doesn't export internal functions, matching sendVerificationEmail's existing precedent) — T27 integration-tests both through the actual upload endpoint. Also adds pricing.js to the tmpdir file-copy list in every test that spawns server.js in isolation — server.js now requires it. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011DUiYs6UjxDMkfvnjtsKFo
-
T20: register now also creates a kind='email_verify' session (24h TTL) and calls sendVerificationEmail() — the one pluggable seam for a real provider once one's chosen (dev-plan-3.md §1 flagged no vendor exists yet). Outside production the register response carries a devVerifyUrl so the whole loop is testable without a real inbox. GET /api/auth/verify?token= flips emailVerified and revokes the token so it can't be replayed. T21: register reads any existing viewer_uid/viewer_nickname cookie (live.html/dance.html's plain client-set identity, unrelated to this session system) and carries it onto the new account as legacyUid/ nickname. Login pushes that legacyUid back onto the browser via Set-Cookie — even from a clean browser with no viewer_uid cookie at all — so logging into an account doesn't let someone dodge per-viewer_uid vote limits by starting a fresh anonymous identity each time. No history is rewritten, per dev-plan-3.md's explicit scope cut. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011Deddn3sM3oQnmvqWSMb4y
-
POST /api/auth/register, POST /api/auth/login, POST /api/auth/logout, and a debug GET /api/auth/me (T19's own acceptance criteria calls for it explicitly). Passwords hashed with node:crypto's scryptSync (salt stored alongside the hash, verified with timingSafeEqual) — still zero new dependencies. Sessions live in a separate user_session cookie from admin_session (T6) — the two are unrelated systems, confirmed by a test that a user session can't unlock /api/admin/*. Login and "email not registered" return the identical 401 body so the endpoint can't be used to enumerate registered emails. Every test file that spawns server.js in an isolated tmpdir now also copies db.js alongside it (dance-channel/dressup-generate/dressup-crud/ admin-pages-flow/uploads-route .test.js) — server.js requires db.js unconditionally now, so those child processes were crashing silently without it. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011Deddn3sM3oQnmvqWSMb4y
