What's new in The Shopping & PMax Growth System

Summary

Agent Codebase 23 → 25 commands. Module 19 (Compounding Growth Loop) extended with pmax-autogrow + creative-refresh-review chapters. URL expansion always OFF on new PMax campaigns. Live mutations dry-run by default.

Highlights

• Agent Codebase — 23 → 25 commands (+2 net-new)
• pmax-autogrow — Weekly PMax self-grow loop — diagnose → mutate → promote → retire, ledger-tracked, dry-run default, one-command rollback
• creative-refresh-review — Scores every new asset refresh against 13-Point Creative v10.5 pre-ship
• URL expansion always OFF — Default on every new PMax campaign; live mutations dry-run with explicit --apply
• 152-point feed quality — Current feed quality requirements preserved, with campaign-type-specific text automation

Full release notes

Scope: Agent Codebase 23 → 25 commands. Blueprint stays at 19 modules (Module 19 Compounding Growth Loop extended with pmax-autogrow + creative-refresh-review chapters).

Net-new commands (+2): pmax-autogrow (weekly PMax self-grow loop — diagnose → mutate → promote → retire, ledger-tracked, dry-run default, one-command rollback), creative-refresh-review (scores every new asset refresh against 13-Point Creative v10.5 before going live).

PMax safety defaults: URL expansion always OFF on every new PMax campaign. Live campaign mutations dry-run by default with explicit --apply required. Campaign-type-specific text automation (PMax-asset-text rules differ from Search RSA rules). Asset content IMMUTABLE; associations MUTABLE. Current 152-point feed quality requirements preserved.

Curriculum updates: 2026-04 Update section added to sales page covering autogrowth + creative review + URL expansion safety. Module 19 chapters extended.

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