Built fast. Shipped honestly.

Every scaffold is now generated by Claude Opus 5, the most capable model available for this kind of work. Same scoping flow, noticeably sharper output. In side-by-side testing on the same project, the new scaffolds trade filler for detail that is actually specific to what you are building: which permissions your app needs and why, the gotcha that bites on step three, a rollback section that reflects how your platform really works. Shorter in places, and worth more per line.

The interview got faster. The questions come back quicker, so scoping a project feels less like waiting and more like talking.

Step away mid-interview and your answers are still there. If a session timed out, a phone backgrounded the tab, or you got pulled into something else, the interview used to start over from scratch. Your progress is now saved as you go and restored when you come back, with a "Progress restored" tag so you know what happened.

Long renders are no longer interrupted. Kicking off a Super Scaffold and switching tabs could cancel a generation that was working perfectly. It runs to completion now.

Every finished project reaches your dashboard. In some cases a scaffold could generate on screen and not show up in your project list afterwards. Fixed, and if a project ever fails to save you will now see a clear warning telling you to download before leaving, rather than finding out later.

Scaffolds are dated correctly. The changelog stub in generated docs carried a placeholder date instead of today's.

Grok joins the build agents. Tell the interview which agent you build with — Claude Code, Codex, OpenClaw, Hermes, or Grok — and your scaffold is written for it (the right primary doc, the right conventions).

Out-of-band uptime monitoring. An independent monitor now watches our daily ops job, so if our email pipeline ever has a problem we hear about it immediately instead of finding out late.

"I build with OpenClaw" no longer turns your app into a skill. OpenClaw and Hermes are coding agents and they have skills; naming one as your build tool used to occasionally flip the scaffold into "build me a skill" mode. Now only the project-type picker decides that.

Transactional email reliability. A server-config issue was silently causing some welcome / onboarding / reminder emails to fail to send. Fixed, hardened so it can't recur, and every affected message was re-sent to the people who missed it.

Project-type picker. Every new project now starts by choosing what you're building: App, Website, OpenClaw skill, or Hermes skill. The whole interview and scaffold tailor to your choice.

Real, complete agent skills. Pick a skill and you get back a full, install-ready folder — SKILL.md with correct frontmatter and step-by-step instructions, plus references/, scripts/, and templates/ — not a thin one-file skill the agent half-ignores.

The interview researches tools it doesn't know. Name any tool, framework, or service and it looks it up and confirms what it found before scoping around it. No more "never heard of it."

Skill-building guides. Free, public step-by-step guides for creating OpenClaw and Hermes skills, at /guides.

First-run walkthrough on the new-project screen for new and returning users.

Home page now showcases what's new, with a live render demo (and no more layout jump while it loops). Skill files show up in the live file list as they generate.

Quick Scope. Describe your project in a single sentence and get a complete scaffold auto-generated — no full interview required. The fastest path from idea to a ready-to-use plan.

Super Scaffolding. For projects bigger than a weekend: instead of the standard four files, get one file per implementation phase (typically 10–14 files), each with the decisions to make, risks to watch, tasks to do, and acceptance criteria — laid out before any code is written. Re-render any existing project as Super from its detail page.

In-app support bubble. Floating "?" bottom-right of every authenticated page. Click → modal with type radio (Bug / Feature / Question / General feedback) + textarea. Submissions email support@aipersonamethod.com from jeff@aiprojectscoper.com with Reply-To set to the submitter (so admin replies route back). Bug reports auto-attach a 5-minute diagnostic blob: console errors, failed fetches, route history, viewport, browser, connection. Admin inbox at /admin/support.

Set / change password card on /dashboard/account. Closes the loop on the new "magic link is the primary sign-in" flow — users who got in passwordless can establish or update a password from inside the app.

Onboarding email sequence (4 emails). Day 1 / 3 / 7 / 14 with essay-style content (project briefs, the four-file pyramid, prompt patterns, multi-session workflows). Daily Vercel cron at 9 AM PT, smart backfill so existing users catch up gradually. Day 1 fired to 43 alpha users today.

Session refresh guard. Mounted in dashboard + admin layouts. Refreshes Supabase access tokens on tab focus + every 10 min while visible. Gracefully bounces to /login?next=... on terminal session loss.

Live Founder seat counter on the landing page. Mirrors the one on /pricing.

Real-revenue MRR / LTV / churn tracking. Replaces the list-price MRR estimate on /admin with three measured variants: Actual MRR (cash collected last cycle), Expected MRR (forward projection at current discount), List MRR (no discount). Discount cost rendered alongside. Adds Total LTV (gross paid minus refunds) and a "biggest customers" leaderboard. New app.invoice_events ledger backs all of it.

Promo code redemption tracking. /admin/promotions now shows who redeemed each code; /admin/users/[id] shows the promo on the active sub; /admin/subscriptions adds a Promo column.

Guardrails middleware on /api/chat + /api/render-plan. Input-side blocks prompt injection / extraction / jailbreak / data-exfiltration attempts. Output-side scans for system-prompt leakage, PII, API keys, internal paths. 18 default rules, /admin/guardrails audit log, 88 vitest tests.

/admin/support, /admin/guardrails admin pages.

Signup flow now skips the email-check step. Previously: fill form → "Check your inbox" → find email → click link → pricing. Most users dropped at the email step (zero of six recent unpaid signups had opened their welcome email). Now: fill form → redirected straight to /pricing, signed in. The welcome email still fires as a backup; it's no longer the only way in.

/login "Email me a sign-in link" is now a primary affordance. Closed-alpha users provisioned via Request Access have a server-generated random password they were never shown; password login fails with "Invalid login credentials." The magic-link button is the right entry point, surfaced prominently with a hint when password attempts fail.

/forgot-password rewired through the working magic-link endpoint. The previous Supabase reset-password flow emitted fragment-token URLs that our /auth/callback couldn't handle.

Magic-link destination routes paid users to /dashboard, unpaid to /pricing. Previously defaulted everyone to /pricing regardless of subscription state.

/admin/coupons link removed from admin nav. The page itself still resolves at the URL for historical inspection; it's no longer promoted in the UI since the redemption pathway was deprecated. Use /admin/promotions for active discount management.

Admin users couldn't render their own product. Admin accounts without a Stripe subscription were treated as free tier (limit = 0 renders), so every Generate Scaffold attempt hit 429 and fell back to a generic template zip. Admins now bypass the per-tier render cap.

Paywall bypass. The chat-view client's error catch built a template-based zip locally on any thrown error, including 429 (rate limited / over cap) and 403 (suspended). So anyone hitting the paywall still walked away with a working starter zip. Closed: 429 and 403 now surface clear "upgrade" / "contact support" messages and skip the template fallback. Other errors (timeouts, partial streams) still degrade gracefully.

"Tailored render failed - building from templates instead" error trace pointed to multiple causes — the admin tier issue above is the dominant one tonight.

Stale test-mode subscription orphaned in app.subscriptions was throwing on every Stripe backfill. Removed.

Magic-link confirmation for vastafferai+aaicc@gmail.com and 5 other affected users via fresh sends from the fixed endpoint after the magic-link routing fix.

/api/auth/magic-link rate-limit cooldown mid-incident (Surbs hit it during the 6-user catch-up; resent via fire-launch-to-leads.mjs script).

Render token cap raised from 8K → 64K with surface-on-cap error so users see truncation instead of mystery silence.

CRON_SECRET env var had a trailing newline causing every Vercel build after the env-add to fail with "leading or trailing whitespace not allowed in HTTP header values." Removed and re-added with printf instead of echo.

40 apology emails to unpaid users about the morning's sign-in friction (response: most ignored, a couple converted).

43 Day 1 onboarding emails triggered manually before the 9 AM PT cron's first scheduled run.

Migrations 006 (guardrail_events), 008 (subscription promo tracking), 009 (revenue / LTV / churn), 010 (support_messages), 011 (support_messages.diagnostics jsonb).

lib/guardrails/, lib/billing/ledger.ts, lib/admin/mrr.ts, lib/support/diagnostics.ts, components/auth/session-refresh-guard.tsx, components/support/support-bubble.tsx, components/support/diagnostic-collector.tsx.

web/scripts/backfill-subscription-promos.mjs, web/scripts/backfill-invoice-history.mjs, web/scripts/fire-apology-to-unpaid.mjs.

vitest 2.1.9 added; 88 tests on guardrails module passing.

Vercel Cron (vercel.json) added for daily onboarding at 16:00 UTC.