- Work only in
D:\unhold-fresh(clean clone of github.com/thribhuvan003/unhold). - Scope: remove real-user confusion on critical path; no full redesign; no auto-post to Reddit/X.
- Product promise stays: organise facts/papers + review-before-send drafts; user sends everything; no unfreeze guarantee.
- Letter system stays template + case-aware slots, optional LLM with fallback (do not market as “AI lawyer”).
- Demo/agents audit showed users can think demo = their case, AI always writes a unique lawyer letter, and skipping AI blocks drafts (false).
- Live site historically over-claimed vs safety contract; copy must match product behaviour before cold traffic.
- Full visual redesign + agent rewrite before launch (too large, delays honest distribution).
- Launch-pack-only without UX fixes (would send confused users into the funnel).
- Auto-posting to Reddit/X without human approval of final text.
messages/en.json+messages/hi.json: home steps, intake hints, consent AI-skip truth, demo truth block, letter unlock copy, package/authority framing, draft-loading jargon.app/[locale]/demo/page.tsx: “truth” list so demo is not misleading.RequestDraft/DraftPendingRefresh: use i18n; honest “preparing draft” language.- Papers page: use translation keys instead of hardcoded English.
- Deploy to unhold.live after review.
- User-approved Reddit/X posts only.
- Optional: e2e smoke after deploy; screenshot refresh under
docs/screenshots.