2026-04 Page Review Panel + Canvas-Design Default

Summary

Agent Codebase 36 → 42 commands. Module 18 added (Page Review Panel + Canvas Design). Net-new: presell-review, offer-review, presell-review-backfill, presell-smoke-test, presell-rollback, brand-voc-mine. Canvas-design from-scratch is now the default on every new build.

Highlights

• Agent Codebase - 36 → 42 commands (+6 net-new)
• Page Review Panel - Scanner + expert reviewer scoring, P0/P1/P2 findings, combined page-health score, trust-level gating
• Canvas-design default - From-scratch on every new build; saved themes opt-in only via theme=
• Smoke test + rollback - presell-smoke-test catches tracking/rendering/CTA failures pre-ship; presell-rollback undoes any deploy in one command
• brand-voc-mine - Harvests buyer language from reviews, Reddit, support logs
• Module 18 added - Page Review Panel + Canvas Design

Full release notes

Scope: Agent Codebase 36 → 42 commands. Blueprint 17 → 18 modules (Module 18: Page Review Panel + Canvas Design).

Net-new commands (+6): presell-review, offer-review, presell-review-backfill, presell-smoke-test, presell-rollback, brand-voc-mine.

Design system change: Canvas-design from-scratch is now the default on every new presell/offer build. Saved themes are opt-in only via theme=<name> or an explicit --auto-match flag. Theme Registry remains available (91 presell + 12 offer + 25 image themes) as a named opt-in, not a default.

Page Review Panel: Two-layer review stack - deterministic scanner + expert reviewer ensemble. Combined page-health score (0-100) gated by trust level. P0 blocks ship; P1 flags rework; P2 to backlog. Every review saved to page-review-history.jsonl and feeds attribution + brand calibration.

Curriculum updates: Module 18 added. Tier-comparison table updated with Canvas-design default, Page Review Panel, smoke test + rollback, brand VOC mining, review backfill rows.

Global framework changes (apply across every product)

• Phase 0 objective brief required on presell, offer, Search, PMax, Demand Gen, and review artifacts before generation begins
• Page Review Panel where relevant - scanner + expert reviewer scoring, P0/P1/P2 findings, combined page-health score, trust-level gating
• PMax safety defaults - URL expansion always OFF, dry-run mutations, explicit --apply required
• Search framework rules - exact-match only, CPC floor/ceiling, IF-function ban, single DKI per RSA, headline composition standards
• Canvas-design default - saved themes opt-in only via theme=<name> or explicit --auto-match flag
• AI image/video framework refresh - campaign brief fields, DR creative standards, creative feedback loops, asset posteriors, multi-aspect video standards
• AutoResearch eval system - review eval buckets added for presell-review and offer-review
• Claude Code + Codex compatibility - CLAUDE.md remains the Claude Code entrypoint; AGENTS.md symlinked for Codex

Security / client-readiness

• No API keys, OAuth secrets, service-account files, .env files, or runtime state files shipped in any codebase
• No source-brand or client names included
• No maintainer-local paths included
• MCP configs use placeholders or environment variables only
• READMEs reflect current command counts and current 2026-04 product positioning