feat(crypto): Phase 2a — declare encryption intent for tags/scenes/missions

Preparation step for the space-scoped data model migration (Phase 2b).
Moves globalTags, tagGroups, workbenchScenes, and aiMissions from the
plaintext allowlist into the encryption registry with enabled:false —
so the audit script documents which fields WILL be encrypted without
changing any runtime behaviour.

Fields chosen per design-doc:
- globalTags.name — personal categorization (Therapie, Finanzen-privat)
- tagGroups.name — same
- workbenchScenes.name + description — scene labels often encode
  Space-specific context (Q2-Launch, Urlaub 2026)
- aiMissions.title + conceptMarkdown + objective — all user-typed
  mission config; state/cadence/inputs stay plaintext for the Runner

Deliberately kept plaintext (against my initial suggestion):
- aiAgents.name — registry comment explains: name is the Actor
  displayName cache key for historic attribution. Encrypting would
  show "🤖 [encrypted]" on every past task the agent ever touched.
- globalTags.icon / tagGroups.icon / color — not personal content;
  icon is a visual cue, color is theme metadata

The 2c migration (Dexie v35, flip enabled:true) runs after 2b lands
the schema changes so existing rows get encrypted in one controlled
pass instead of mixing schema + encryption in the same upgrade.

Crypto audit: 195 Dexie tables classified (94 encrypted, 101
plaintext-allowlisted). Type-check clean.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

apps/mana/apps/web/src/lib/data/crypto/plaintext-allowlist.ts

@@ -19,7 +19,6 @@
 export const PLAINTEXT_ALLOWLIST: readonly string[] = [
 	'achievements', // TODO: audit
 	'activities', // TODO: audit
-	'aiMissions', // TODO: audit
 	'albumItems', // TODO: audit
 	'albums', // TODO: audit
 	'articleTags', // FK-only junction into globalTags (articleId, tagId). Tag names live in globalTags.
@@ -50,7 +49,6 @@ export const PLAINTEXT_ALLOWLIST: readonly string[] = [
 	'fileTags', // TODO: audit
 	'financeCategories', // TODO: audit
 	'foodFavorites', // TODO: audit
-	'globalTags', // TODO: audit
 	'goals', // TODO: audit
 	'guideCollections', // TODO: audit
 	'guideTags', // TODO: audit
@@ -99,7 +97,6 @@ export const PLAINTEXT_ALLOWLIST: readonly string[] = [
 	'songTags', // TODO: audit
 	'spaceMembers', // TODO: audit
 	'storageFolders', // TODO: audit
-	'tagGroups', // TODO: audit
 	'taskLabels', // TODO: audit
 	'timeAlarms', // TODO: audit
 	'timeBlockTags', // TODO: audit
@@ -121,5 +118,4 @@ export const PLAINTEXT_ALLOWLIST: readonly string[] = [
 	'wishesItems', // TODO: audit
 	'wishesLists', // TODO: audit
 	'wishesPriceChecks', // TODO: audit
-	'workbenchScenes', // TODO: audit
 ];

apps/mana/apps/web/src/lib/data/crypto/registry.ts

@@ -573,6 +573,42 @@ export const ENCRYPTION_REGISTRY: Record<string, EncryptionConfig> = {
 	// structural fields.
 	agents: { enabled: true, fields: ['systemPrompt', 'memory'] },
 
+	// ─── AI Missions ─────────────────────────────────────────
+	// docs/plans/space-scoped-data-model.md §2a — declared with
+	// enabled:false during prep so the audit script is happy; flips to
+	// true in 2c alongside the Dexie v35 encryption migration.
+	//
+	// User-typed content on missions: `title` (display label the user
+	// types at create time), `conceptMarkdown` (free-form context the
+	// planner reads), `objective` (the actionable goal string). State,
+	// cadence, inputs (FK-only), nextRunAt, iterations, agentId all
+	// stay plaintext — needed for the Runner's "due now" index walk
+	// and mission-detail filters.
+	aiMissions: { enabled: false, fields: ['title', 'conceptMarkdown', 'objective'] },
+
+	// ─── Tags (shared-stores) ────────────────────────────────
+	// docs/plans/space-scoped-data-model.md §2a — declared with
+	// enabled:false during prep; flips to true in 2c. Tag names like
+	// "Therapie" or "Finanzen-privat" can leak personal categorization,
+	// so they belong under encryption. `color` + `icon` + `groupId` +
+	// `sortOrder` stay plaintext: they're visual metadata + the group
+	// FK, none of which leak sensitive taxonomy. `name` is NOT indexed
+	// for .where() lookups today, so encrypting it is safe — dedupe-
+	// within-space lookups go through the new [spaceId+name] index after
+	// Phase 2b and run over already-decrypted rows in the scoped store.
+	globalTags: { enabled: false, fields: ['name'] },
+	tagGroups: { enabled: false, fields: ['name'] },
+
+	// ─── Workbench Scenes ────────────────────────────────────
+	// docs/plans/space-scoped-data-model.md §2a — declared with
+	// enabled:false during prep; flips to true in 2c. `name` is the
+	// user-visible scene label ("Heute", "Q2-Launch") and `description`
+	// is the short subtitle — both are user-typed free text that can
+	// leak Space-specific context. openApps / order / wallpaper /
+	// viewingAsAgentId / scopeTagIds stay plaintext (structural /
+	// indexed / foreign-key data).
+	workbenchScenes: { enabled: false, fields: ['name', 'description'] },
+
 	// ─── Articles (Pocket-style read-it-later) ──────────────
 	// Reading-behaviour data — same sensitivity class as newsArticles.
 	// Encrypted: