console: agent-first mode (console plan / console verify), verified App Privacy and Play Create app facts

- `storebound console plan <form> [--json] [--lookup]`: read-only plan for an AI agent that drives the browser
  itself: start URL (deep links), ordered steps, answer sheet with the console's exact labels (certain / ask owner,
  curly quotes noted), never-press list, verified UI facts, driving guidance (in-page async loop + window.__log),
  verify instructions. All six forms.
- `storebound console verify <form> [--from-text <file|->]`: App Privacy per-type summaries from page text, Play
  Create app from input states, content rating / target audience summaries, App Store record and age rating from
  the API; exit 1 on a mismatch.
- MCP: console_plan, console_verify (read-only); console fill stays off MCP.
- App Privacy (verified 2026-10-12): two information steps, unlabeled purpose checkboxes (nearest parent text),
  "Definitions and Examples" never pressed; fixture and engine follow.
- Play Create app (verified 2026-10-12): /developers/<id>/create-new-app, App package name + read-only Check
  availability, material-radio inner [role=radio] click, app dashboard URL and play.appId, service account
  "Add app" as an owner step.
- console-autofill skill rewritten agent-first; agents-core rule 10; README; pitfalls 12, 18 (confirmed sign-out),
  19; CHANGELOG; play-first-release.

tests/growth-config.test.ts

@@ -33,8 +33,8 @@ describe('growth extras config', () => {
 });
 
 describe('growth extras over MCP', () => {
-  test('no console tool exists in MCP', () => {
-    expect(TOOLS.filter(t => t.command[0] === 'console' || /console/.test(t.name))).toEqual([]);
+  test('no console tool that drives a browser exists in MCP (only the read-only plan and verify)', () => {
+    expect(TOOLS.filter(t => (t.command[0] === 'console' || /console/.test(t.name)) && !(t.write === false && ['plan', 'verify'].includes(t.command[1] ?? ''))).map(t => t.name)).toEqual([]);
   });
   test('Apple Ads, RevenueCat, device registration and internal sharing are read-only in MCP', () => {
     const risky = TOOLS.filter(t => (t.command[0] === 'ads' && t.command[1] === 'apple') || t.command[0] === 'revenuecat' || (t.command[0] === 'play' && t.command[1] === 'share')