diff --git a/.claude/commands/improve-codebase-architecture.md b/.claude/commands/improve-codebase-architecture.md
new file mode 100644
index 00000000..cfb69c76
--- /dev/null
+++ b/.claude/commands/improve-codebase-architecture.md
@@ -0,0 +1,11 @@
+---
+description: Matt Pocock의 improve-codebase-architecture — 코드베이스 deepening opportunities 발견, 도메인 언어·ADR 기반 refactoring 제안
+---
+
+Read `~/.agents/skills/improve-codebase-architecture/SKILL.md` and follow its instructions exactly. Treat the loaded SKILL.md content as executable instructions, not reference material.
+
+Before exploring, read the domain docs as listed in [`docs/agents/domain.md`](docs/agents/domain.md) and any relevant ADRs in `docs/adr/`. Use `subagent_type=Explore` for codebase walks.
+
+The user's target area to improve: $ARGUMENTS
+
+If `$ARGUMENTS` is empty, ask which area/package to scope (e.g. `packages/web/lib/supabase`, `packages/api-server`, design-system) before exploring. Don't run on the entire monorepo without scope.
diff --git a/.gitignore b/.gitignore
index f971d556..bd281c2b 100644
--- a/.gitignore
+++ b/.gitignore
@@ -85,6 +85,15 @@ packages/ai-server/searxng/settings.yml.new
 !.omc/
 !.omc/project-memory.json
 
+# OMC per-package artifacts (root .omc 정책은 위에서 처리)
+packages/*/.omc/
+
+# Next.js build backups
+packages/*/.next.bak/
+
+# Local scratch directory
+.scratch/
+
 # Hybrid harness local artifacts
 packages/web/.tasks/
 packages/web/.handoff/
diff --git a/docs/superpowers/plans/2026-05-07-profile-tries-detail-modal.md b/docs/superpowers/plans/2026-05-07-profile-tries-detail-modal.md
new file mode 100644
index 00000000..5ec91693
--- /dev/null
+++ b/docs/superpowers/plans/2026-05-07-profile-tries-detail-modal.md
@@ -0,0 +1,1293 @@
+# Profile Tries Detail Modal Implementation Plan
+
+> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
+
+**Goal:** Clicking a saved try-on in `/profile` -> `Tries` opens an in-place detail modal backed by save-time provenance snapshots and safe legacy fallbacks.
+
+**Architecture:** Keep the existing `user_tryon_history` table shape. Expand the POST save payload into `style_combination`, normalize that JSON in the profile tries route, regenerate the generated web API model from the OpenAPI source, and keep the detail UI local to the profile grid.
+
+**Tech Stack:** Next.js App Router route handlers, Supabase JS, React 19, TanStack Query, Vitest, Testing Library, Orval generated API models.
+
+---
+
+## Spec Review Gate
+
+Verdict: proceed to implementation planning.
+
+Required planning adjustment: the spec's file list must include the OpenAPI source path because `TriesGrid` imports `TryItem` from `packages/web/lib/api/generated/models`. Do not hand-edit `packages/web/lib/api/generated/*`; update the source schema in `packages/api-server/src/domains/users/dto.rs`, regenerate `packages/api-server/openapi.json` through the repo's API-server flow if needed, then run `bun run generate:api` from `packages/web`.
+
+No migration is part of this pass. Storage migration remains a follow-up.
+
+## File Structure
+
+- Modify `packages/web/app/api/v1/tries/route.ts`
+  - Validate expanded save payload fields.
+  - Store `person_original_image`, `source_post_snapshot`, and `selected_items_snapshot` inside `style_combination`.
+- Modify `packages/web/app/api/v1/tries/__tests__/route.test.ts`
+  - Cover snapshot save and legacy ID-only save.
+- Modify `packages/web/app/api/v1/users/me/tries/route.ts`
+  - Select `style_combination`.
+  - Normalize malformed, null, legacy, and snapshot rows.
+  - Resolve current post and solution rows only for legacy rows with IDs and no snapshots.
+- Modify `packages/web/app/api/v1/users/me/tries/__tests__/route.test.ts`
+  - Cover snapshot response and legacy fallback response.
+- Modify `packages/api-server/src/domains/users/dto.rs`
+  - Expand `TryItem` schema with nullable source post, selected item, and original image fields.
+- Regenerate `packages/api-server/openapi.json` if this repo's API-server command is available.
+- Regenerate `packages/web/lib/api/generated/models/tryItem.ts` and related generated files via `bun run generate:api`.
+- Modify `packages/web/lib/hooks/useVtonTryOn.ts`
+  - Accept source post snapshot and include snapshots in save request.
+- Modify `packages/web/lib/components/vton/VtonModal.tsx`
+  - Derive `sourcePostSnapshot` from `posts` and pass it to the hook.
+- Create `packages/web/lib/components/profile/TryDetailModal.tsx`
+  - Present detail-ready try item content.
+- Modify `packages/web/lib/components/profile/TriesGrid.tsx`
+  - Own selected try state and render `TryDetailModal`.
+- Create `packages/web/lib/components/profile/__tests__/TryDetailModal.test.tsx`
+  - Verify full metadata and fallback rendering.
+- Create `packages/web/lib/components/profile/__tests__/TriesGrid.test.tsx`
+  - Verify card click opens the modal.
+
+## Task 1: Save Expanded Snapshot Contract
+
+**Files:**
+- Modify: `packages/web/app/api/v1/tries/route.ts`
+- Modify: `packages/web/app/api/v1/tries/__tests__/route.test.ts`
+
+- [ ] **Step 1: Add failing POST snapshot test**
+
+Add this test case to `describe("POST /api/v1/tries", ...)` in `packages/web/app/api/v1/tries/__tests__/route.test.ts`.
+
+```ts
+it("stores detail snapshots in style_combination", async () => {
+  getUserMock.mockResolvedValue({ data: { user: { id: "user-1" } } });
+  singleMock.mockResolvedValue({
+    data: {
+      id: "try-1",
+      image_url: "data:image/png;base64,result",
+      created_at: "2026-05-07T00:00:00Z",
+    },
+    error: null,
+  });
+
+  const { POST } = await import("../route");
+  const res = await POST(
+    makeRequest({
+      result_image: "data:image/png;base64,result",
+      person_original_image: "data:image/png;base64,person",
+      source_post_id: "post-1",
+      selected_item_ids: ["item-1", "item-2"],
+      source_post_snapshot: {
+        id: "post-1",
+        title: "Source Look",
+        image_url: "https://example.com/post.jpg",
+        artist_name: "Artist",
+        group_name: null,
+        context: "Airport look",
+      },
+      selected_items_snapshot: [
+        {
+          id: "item-1",
+          title: "Jacket",
+          thumbnail_url: "https://example.com/jacket.jpg",
+          description: "Black jacket",
+          keywords: ["outerwear"],
+        },
+      ],
+    })
+  );
+
+  expect(res.status).toBe(201);
+  expect(insertMock).toHaveBeenCalledWith("user_tryon_history", {
+    user_id: "user-1",
+    image_url: "data:image/png;base64,result",
+    style_combination: {
+      source_post_id: "post-1",
+      selected_item_ids: ["item-1", "item-2"],
+      person_original_image: "data:image/png;base64,person",
+      source_post_snapshot: {
+        id: "post-1",
+        title: "Source Look",
+        image_url: "https://example.com/post.jpg",
+        artist_name: "Artist",
+        group_name: null,
+        context: "Airport look",
+      },
+      selected_items_snapshot: [
+        {
+          id: "item-1",
+          title: "Jacket",
+          thumbnail_url: "https://example.com/jacket.jpg",
+          description: "Black jacket",
+          keywords: ["outerwear"],
+        },
+      ],
+    },
+  });
+});
+```
+
+- [ ] **Step 2: Run the failing POST test**
+
+Run:
+
+```bash
+cd packages/web
+bun run test:unit app/api/v1/tries/__tests__/route.test.ts
+```
+
+Expected: the new test fails because `person_original_image`, `source_post_snapshot`, and `selected_items_snapshot` are not stored.
+
+- [ ] **Step 3: Implement payload normalization**
+
+In `packages/web/app/api/v1/tries/route.ts`, replace `SaveTryOnBody` and add the helpers near `toStringArray`.
+
+```ts
+interface SourcePostSnapshot {
+  id: string;
+  title: string | null;
+  image_url: string | null;
+  artist_name?: string | null;
+  group_name?: string | null;
+  context?: string | null;
+}
+
+interface SelectedItemSnapshot {
+  id: string;
+  title: string;
+  thumbnail_url: string;
+  description?: string | null;
+  keywords?: string[] | null;
+}
+
+interface SaveTryOnBody {
+  result_image?: unknown;
+  image_url?: unknown;
+  person_original_image?: unknown;
+  source_post_id?: unknown;
+  selected_item_ids?: unknown;
+  source_post_snapshot?: unknown;
+  selected_items_snapshot?: unknown;
+}
+
+function optionalString(value: unknown): string | null {
+  return typeof value === "string" && value.length > 0 ? value : null;
+}
+
+function nullableString(value: unknown): string | null {
+  return typeof value === "string" ? value : null;
+}
+
+function toSourcePostSnapshot(value: unknown): SourcePostSnapshot | null {
+  if (!value || typeof value !== "object") return null;
+  const input = value as Record<string, unknown>;
+  const id = optionalString(input.id);
+  if (!id) return null;
+  return {
+    id,
+    title: nullableString(input.title),
+    image_url: nullableString(input.image_url),
+    artist_name: nullableString(input.artist_name),
+    group_name: nullableString(input.group_name),
+    context: nullableString(input.context),
+  };
+}
+
+function toSelectedItemSnapshots(value: unknown): SelectedItemSnapshot[] {
+  if (!Array.isArray(value)) return [];
+  return value.flatMap((entry) => {
+    if (!entry || typeof entry !== "object") return [];
+    const input = entry as Record<string, unknown>;
+    const id = optionalString(input.id);
+    const title = optionalString(input.title);
+    const thumbnailUrl = optionalString(input.thumbnail_url);
+    if (!id || !title || !thumbnailUrl) return [];
+    return [
+      {
+        id,
+        title,
+        thumbnail_url: thumbnailUrl,
+        description: nullableString(input.description),
+        keywords: Array.isArray(input.keywords)
+          ? input.keywords.filter(
+              (keyword): keyword is string => typeof keyword === "string"
+            )
+          : null,
+      },
+    ];
+  });
+}
+```
+
+Then update the `style_combination` insert object.
+
+```ts
+const styleCombination = {
+  source_post_id:
+    typeof body.source_post_id === "string" ? body.source_post_id : null,
+  selected_item_ids: toStringArray(body.selected_item_ids),
+  person_original_image: optionalString(body.person_original_image),
+  source_post_snapshot: toSourcePostSnapshot(body.source_post_snapshot),
+  selected_items_snapshot: toSelectedItemSnapshots(
+    body.selected_items_snapshot
+  ),
+};
+```
+
+Use `style_combination: styleCombination` in the insert.
+
+- [ ] **Step 4: Keep legacy test green**
+
+Update existing expectations in `route.test.ts` for legacy payloads to include the newly persisted nullable fields.
+
+```ts
+style_combination: {
+  source_post_id: "post-1",
+  selected_item_ids: ["item-1", "item-2"],
+  person_original_image: null,
+  source_post_snapshot: null,
+  selected_items_snapshot: [],
+},
+```
+
+- [ ] **Step 5: Verify POST route tests**
+
+Run:
+
+```bash
+cd packages/web
+bun run test:unit app/api/v1/tries/__tests__/route.test.ts
+```
+
+Expected: all tests in `route.test.ts` pass.
+
+## Task 2: Return Detail-Ready Try Items
+
+**Files:**
+- Modify: `packages/web/app/api/v1/users/me/tries/route.ts`
+- Modify: `packages/web/app/api/v1/users/me/tries/__tests__/route.test.ts`
+
+- [ ] **Step 1: Add failing snapshot response test**
+
+Extend the mock chain in `route.test.ts` so `.from("user_tryon_history")` can return the existing chain, then add this test.
+
+```ts
+it("returns expanded fields from saved snapshots", async () => {
+  getUserMock.mockResolvedValue({ data: { user: { id: "user-1" } } });
+  rangeMock.mockResolvedValue({
+    data: [
+      {
+        id: "try-1",
+        image_url: "https://example.com/result.png",
+        created_at: "2026-05-07T00:00:00Z",
+        style_combination: {
+          source_post_id: "post-1",
+          selected_item_ids: ["item-1"],
+          person_original_image: "https://example.com/person.png",
+          source_post_snapshot: {
+            id: "post-1",
+            title: "Source Look",
+            image_url: "https://example.com/post.png",
+            artist_name: "Artist",
+            group_name: null,
+            context: "Stage look",
+          },
+          selected_items_snapshot: [
+            {
+              id: "item-1",
+              title: "Jacket",
+              thumbnail_url: "https://example.com/jacket.png",
+              description: "Black jacket",
+              keywords: ["outerwear"],
+            },
+          ],
+        },
+      },
+    ],
+    error: null,
+    count: 1,
+  });
+
+  const { GET } = await import("../route");
+  const res = await GET(makeRequest());
+  const json = await res.json();
+
+  expect(res.status).toBe(200);
+  expect(selectMock).toHaveBeenCalledWith(
+    "user_tryon_history",
+    "id, image_url, created_at, style_combination",
+    { count: "exact" }
+  );
+  expect(json.data[0]).toEqual({
+    id: "try-1",
+    image_url: "https://example.com/result.png",
+    created_at: "2026-05-07T00:00:00Z",
+    source_post_id: "post-1",
+    selected_item_ids: ["item-1"],
+    person_original_image: "https://example.com/person.png",
+    source_post: {
+      id: "post-1",
+      title: "Source Look",
+      image_url: "https://example.com/post.png",
+      artist_name: "Artist",
+      group_name: null,
+      context: "Stage look",
+    },
+    selected_items: [
+      {
+        id: "item-1",
+        title: "Jacket",
+        thumbnail_url: "https://example.com/jacket.png",
+        description: "Black jacket",
+        keywords: ["outerwear"],
+      },
+    ],
+  });
+});
+```
+
+- [ ] **Step 2: Add failing legacy fallback test**
+
+Add a test where `style_combination` is null. It should return detail fields with nulls and empty arrays.
+
+```ts
+it("returns fallback detail fields for null style_combination", async () => {
+  getUserMock.mockResolvedValue({ data: { user: { id: "user-1" } } });
+  rangeMock.mockResolvedValue({
+    data: [
+      {
+        id: "try-legacy",
+        image_url: "https://example.com/legacy.png",
+        created_at: "2026-05-07T00:00:00Z",
+        style_combination: null,
+      },
+    ],
+    error: null,
+    count: 1,
+  });
+
+  const { GET } = await import("../route");
+  const res = await GET(makeRequest());
+  const json = await res.json();
+
+  expect(json.data[0]).toMatchObject({
+    id: "try-legacy",
+    source_post_id: null,
+    selected_item_ids: [],
+    person_original_image: null,
+    source_post: null,
+    selected_items: [],
+  });
+});
+```
+
+- [ ] **Step 3: Run the failing GET route tests**
+
+Run:
+
+```bash
+cd packages/web
+bun run test:unit app/api/v1/users/me/tries/__tests__/route.test.ts
+```
+
+Expected: tests fail because the route does not select or map `style_combination`.
+
+- [ ] **Step 4: Implement normalizers**
+
+Add these types and helpers in `packages/web/app/api/v1/users/me/tries/route.ts`.
+
+```ts
+type TrySourcePost = {
+  id: string;
+  title: string | null;
+  image_url: string | null;
+  artist_name?: string | null;
+  group_name?: string | null;
+  context?: string | null;
+};
+
+type TrySelectedItem = {
+  id: string;
+  title: string;
+  thumbnail_url: string;
+  description?: string | null;
+  keywords?: string[] | null;
+};
+
+type StyleCombination = {
+  source_post_id: string | null;
+  selected_item_ids: string[];
+  person_original_image: string | null;
+  source_post_snapshot: TrySourcePost | null;
+  selected_items_snapshot: TrySelectedItem[];
+};
+
+function stringOrNull(value: unknown): string | null {
+  return typeof value === "string" && value.length > 0 ? value : null;
+}
+
+function stringArray(value: unknown): string[] {
+  if (!Array.isArray(value)) return [];
+  return value.filter((item): item is string => typeof item === "string");
+}
+
+function normalizeSourcePost(value: unknown): TrySourcePost | null {
+  if (!value || typeof value !== "object") return null;
+  const input = value as Record<string, unknown>;
+  const id = stringOrNull(input.id);
+  if (!id) return null;
+  return {
+    id,
+    title: stringOrNull(input.title),
+    image_url: stringOrNull(input.image_url),
+    artist_name: stringOrNull(input.artist_name),
+    group_name: stringOrNull(input.group_name),
+    context: stringOrNull(input.context),
+  };
+}
+
+function normalizeSelectedItems(value: unknown): TrySelectedItem[] {
+  if (!Array.isArray(value)) return [];
+  return value.flatMap((entry) => {
+    if (!entry || typeof entry !== "object") return [];
+    const input = entry as Record<string, unknown>;
+    const id = stringOrNull(input.id);
+    const title = stringOrNull(input.title);
+    const thumbnailUrl = stringOrNull(input.thumbnail_url);
+    if (!id || !title || !thumbnailUrl) return [];
+    return [
+      {
+        id,
+        title,
+        thumbnail_url: thumbnailUrl,
+        description: stringOrNull(input.description),
+        keywords: Array.isArray(input.keywords)
+          ? input.keywords.filter(
+              (keyword): keyword is string => typeof keyword === "string"
+            )
+          : null,
+      },
+    ];
+  });
+}
+
+function normalizeStyleCombination(value: unknown): StyleCombination {
+  if (!value || typeof value !== "object") {
+    return {
+      source_post_id: null,
+      selected_item_ids: [],
+      person_original_image: null,
+      source_post_snapshot: null,
+      selected_items_snapshot: [],
+    };
+  }
+  const input = value as Record<string, unknown>;
+  return {
+    source_post_id: stringOrNull(input.source_post_id),
+    selected_item_ids: stringArray(input.selected_item_ids),
+    person_original_image: stringOrNull(input.person_original_image),
+    source_post_snapshot: normalizeSourcePost(input.source_post_snapshot),
+    selected_items_snapshot: normalizeSelectedItems(
+      input.selected_items_snapshot
+    ),
+  };
+}
+```
+
+- [ ] **Step 5: Select and map detail fields**
+
+Change the Supabase select and response map.
+
+```ts
+.select("id, image_url, created_at, style_combination", { count: "exact" })
+```
+
+Then map rows before returning.
+
+```ts
+const tries = (data ?? []).map((row) => {
+  const style = normalizeStyleCombination(
+    (row as { style_combination?: unknown }).style_combination
+  );
+  return {
+    id: row.id,
+    image_url: row.image_url,
+    created_at: row.created_at,
+    source_post_id: style.source_post_id,
+    selected_item_ids: style.selected_item_ids,
+    person_original_image: style.person_original_image,
+    source_post: style.source_post_snapshot,
+    selected_items: style.selected_items_snapshot,
+  };
+});
+```
+
+Use `data: tries` in `NextResponse.json`.
+
+- [ ] **Step 6: Add best-effort ID lookup for legacy rows**
+
+After the snapshot-only tests pass, add one more test for a legacy row with `source_post_id` and `selected_item_ids` but no snapshots. Mock additional `from("posts")` and `from("solutions")` chains that return current rows. Implement lookup only for missing snapshots, using these select shapes:
+
+```ts
+const { data: postRows } = await supabase
+  .from("posts")
+  .select("id, title, image_url, artist_name, group_name, context")
+  .in("id", sourcePostIds);
+
+const { data: solutionRows } = await supabase
+  .from("solutions")
+  .select("id, title, thumbnail_url, description, keywords")
+  .in("id", selectedItemIds);
+```
+
+Use saved snapshots first, then maps from current rows, then null or `[]`.
+
+- [ ] **Step 7: Verify GET route tests**
+
+Run:
+
+```bash
+cd packages/web
+bun run test:unit app/api/v1/users/me/tries/__tests__/route.test.ts
+```
+
+Expected: all tests pass.
+
+## Task 3: Regenerate TryItem API Model
+
+**Files:**
+- Modify: `packages/api-server/src/domains/users/dto.rs`
+- Regenerate: `packages/api-server/openapi.json`
+- Regenerate: `packages/web/lib/api/generated/models/tryItem.ts`
+- Regenerate: related generated files under `packages/web/lib/api/generated/`
+
+- [ ] **Step 1: Expand OpenAPI DTO source**
+
+In `packages/api-server/src/domains/users/dto.rs`, add nested DTOs near `TryItem` and replace `TryItem`.
+
+```rust
+/// VTON 히스토리 원본 포스트 스냅샷
+#[derive(Debug, Clone, Serialize, Deserialize, ToSchema)]
+pub struct TrySourcePost {
+    pub id: Uuid,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub title: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub image_url: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub artist_name: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub group_name: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub context: Option<String>,
+}
+
+/// VTON 히스토리 선택 아이템 스냅샷
+#[derive(Debug, Clone, Serialize, Deserialize, ToSchema)]
+pub struct TrySelectedItem {
+    pub id: Uuid,
+    pub title: String,
+    pub thumbnail_url: String,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub description: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub keywords: Option<Vec<String>>,
+}
+
+/// VTON 히스토리 아이템
+#[derive(Debug, Clone, Serialize, Deserialize, ToSchema)]
+pub struct TryItem {
+    pub id: Uuid,
+    pub image_url: String,
+    pub created_at: DateTime<Utc>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub source_post_id: Option<Uuid>,
+    pub selected_item_ids: Vec<Uuid>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub person_original_image: Option<String>,
+    #[serde(skip_serializing_if = "Option::is_none")]
+    pub source_post: Option<TrySourcePost>,
+    pub selected_items: Vec<TrySelectedItem>,
+}
+```
+
+- [ ] **Step 2: Keep API-server compile errors explicit**
+
+Run from repo root:
+
+```bash
+cargo check -p api-server
+```
+
+Expected: if `users/service.rs` still constructs the old three-field `TryItem`, Rust reports missing fields.
+
+- [ ] **Step 3: Update API-server service construction**
+
+In `packages/api-server/src/domains/users/service.rs`, update the existing `TryItem` construction to compile with empty detail fields. The Next route owns the detail implementation for this pass.
+
+```rust
+TryItem {
+    id: r.id,
+    image_url: r.image_url,
+    created_at: r.created_at.into(),
+    source_post_id: None,
+    selected_item_ids: Vec::new(),
+    person_original_image: None,
+    source_post: None,
+    selected_items: Vec::new(),
+}
+```
+
+- [ ] **Step 4: Regenerate OpenAPI and web client**
+
+Use the repo's existing OpenAPI dump command if available. If no package script exists, run the local binary command used by this repository.
+
+```bash
+cd packages/api-server
+cargo run --bin dump_openapi
+cd ../web
+bun run generate:api
+```
+
+Expected: `packages/web/lib/api/generated/models/tryItem.ts` includes `source_post_id`, `selected_item_ids`, `person_original_image`, `source_post`, and `selected_items`.
+
+- [ ] **Step 5: Verify generated model shape**
+
+Run:
+
+```bash
+rg -n "source_post_id|selected_items|person_original_image" packages/web/lib/api/generated/models/tryItem.ts
+```
+
+Expected: all three fields are present.
+
+## Task 4: Send Save-Time Snapshots from VTON
+
+**Files:**
+- Modify: `packages/web/lib/hooks/useVtonTryOn.ts`
+- Modify: `packages/web/lib/components/vton/VtonModal.tsx`
+
+- [ ] **Step 1: Extend hook options**
+
+In `useVtonTryOn.ts`, import `VtonPostData` and add snapshot fields.
+
+```ts
+import type { VtonPostData } from "@/lib/hooks/useVtonPostFetch";
+
+interface UseVtonTryOnOptions {
+  personImage: string | null;
+  personPreview: string | null;
+  selectedItems: ItemData[];
+  sourcePostId: string | null;
+  sourcePostSnapshot: Omit<VtonPostData, "items"> | null;
+  displayResultImage: string | null;
+  abortControllerRef: React.RefObject<AbortController | null>;
+  onTryOnStart: () => void;
+  onTryOnComplete: (resultDataUrl: string, latencyMs: number | null) => void;
+  onTryOnError: (message: string) => void;
+  onTryOnFinally: () => void;
+  startBackgroundJob: (
+    personPreview: string,
+    personImageBase64: string,
+    selectedItems: ItemData[]
+  ) => string;
+}
+```
+
+- [ ] **Step 2: Add snapshot payload to save request**
+
+Include these fields in the `JSON.stringify` body inside `handleSaveToProfile`.
+
+```ts
+body: JSON.stringify({
+  result_image: displayResultImage,
+  person_original_image: personPreview,
+  source_post_id: sourcePostId,
+  selected_item_ids: selectedItems.map((i) => i.id),
+  source_post_snapshot: sourcePostSnapshot,
+  selected_items_snapshot: selectedItems.map((item) => ({
+    id: item.id,
+    title: item.title,
+    thumbnail_url: item.thumbnail_url,
+    description: item.description,
+    keywords: item.keywords,
+  })),
+}),
+```
+
+Add `personPreview` and `sourcePostSnapshot` to the callback dependency list.
+
+- [ ] **Step 3: Derive source post snapshot in VtonModal**
+
+In `VtonModal.tsx`, derive a snapshot after `posts` and `sourcePostId` are available.
+
+```ts
+const sourcePostSnapshot = useMemo(() => {
+  if (!sourcePostId) return null;
+  const post = posts.find((entry) => entry.id === sourcePostId);
+  if (!post) return null;
+  return {
+    id: post.id,
+    title: post.title,
+    image_url: post.image_url,
+    artist_name: post.artist_name,
+    group_name: post.group_name,
+    context: post.context,
+  };
+}, [posts, sourcePostId]);
+```
+
+Pass it to `useVtonTryOn`.
+
+```ts
+sourcePostSnapshot,
+```
+
+- [ ] **Step 4: Verify TypeScript for VTON files**
+
+Run:
+
+```bash
+cd packages/web
+bun run typecheck
+```
+
+Expected: no type errors from `useVtonTryOn.ts` or `VtonModal.tsx`.
+
+## Task 5: Build Try Detail Modal
+
+**Files:**
+- Create: `packages/web/lib/components/profile/TryDetailModal.tsx`
+- Create: `packages/web/lib/components/profile/__tests__/TryDetailModal.test.tsx`
+
+- [ ] **Step 1: Write full metadata modal test**
+
+Create `TryDetailModal.test.tsx`.
+
+```tsx
+/**
+ * @vitest-environment jsdom
+ */
+import React from "react";
+import { describe, expect, it, vi } from "vitest";
+import { fireEvent, render, screen } from "@testing-library/react";
+import "@testing-library/jest-dom";
+import { TryDetailModal } from "../TryDetailModal";
+import type { TryItem } from "@/lib/api/generated/models";
+
+vi.mock("next/image", () => ({
+  default: (props: React.ImgHTMLAttributes<HTMLImageElement>) => (
+    <img {...props} />
+  ),
+}));
+
+const fullTry: TryItem = {
+  id: "try-1",
+  image_url: "https://example.com/result.png",
+  created_at: "2026-05-07T00:00:00Z",
+  source_post_id: "post-1",
+  selected_item_ids: ["item-1"],
+  person_original_image: "https://example.com/person.png",
+  source_post: {
+    id: "post-1",
+    title: "Source Look",
+    image_url: "https://example.com/post.png",
+    artist_name: "Artist",
+    group_name: null,
+    context: "Stage look",
+  },
+  selected_items: [
+    {
+      id: "item-1",
+      title: "Jacket",
+      thumbnail_url: "https://example.com/jacket.png",
+      description: "Black jacket",
+      keywords: ["outerwear"],
+    },
+  ],
+};
+
+describe("TryDetailModal", () => {
+  it("renders result, original, source post, and selected items", () => {
+    render(<TryDetailModal tryItem={fullTry} onClose={() => {}} />);
+
+    expect(screen.getByRole("dialog")).toBeInTheDocument();
+    expect(screen.getByAltText("Try-on result")).toBeInTheDocument();
+    expect(screen.getByAltText("Uploaded original")).toBeInTheDocument();
+    expect(screen.getByText("Source Look")).toBeInTheDocument();
+    expect(screen.getByText("Jacket")).toBeInTheDocument();
+    expect(screen.getByText("outerwear")).toBeInTheDocument();
+  });
+
+  it("calls onClose from the close button", () => {
+    const onClose = vi.fn();
+    render(<TryDetailModal tryItem={fullTry} onClose={onClose} />);
+    fireEvent.click(screen.getByRole("button", { name: /close/i }));
+    expect(onClose).toHaveBeenCalled();
+  });
+});
+```
+
+- [ ] **Step 2: Add fallback modal test**
+
+Add this case to the same test file.
+
+```tsx
+it("renders fallback text when metadata is missing", () => {
+  render(
+    <TryDetailModal
+      tryItem={{
+        id: "try-legacy",
+        image_url: "https://example.com/result.png",
+        created_at: "2026-05-07T00:00:00Z",
+        source_post_id: null,
+        selected_item_ids: [],
+        person_original_image: null,
+        source_post: null,
+        selected_items: [],
+      }}
+      onClose={() => {}}
+    />
+  );
+
+  expect(screen.getByText("Original image unavailable")).toBeInTheDocument();
+  expect(screen.getByText("Source post unavailable")).toBeInTheDocument();
+  expect(screen.getByText("Items unavailable")).toBeInTheDocument();
+});
+```
+
+- [ ] **Step 3: Run failing modal tests**
+
+Run:
+
+```bash
+cd packages/web
+bun run test:unit lib/components/profile/__tests__/TryDetailModal.test.tsx
+```
+
+Expected: module not found until `TryDetailModal.tsx` is created.
+
+- [ ] **Step 4: Implement TryDetailModal**
+
+Create `packages/web/lib/components/profile/TryDetailModal.tsx`.
+
+```tsx
+"use client";
+
+import { useEffect } from "react";
+import Image from "next/image";
+import Link from "next/link";
+import { X } from "lucide-react";
+import type { TryItem } from "@/lib/api/generated/models";
+
+interface TryDetailModalProps {
+  tryItem: TryItem;
+  onClose: () => void;
+}
+
+export function TryDetailModal({ tryItem, onClose }: TryDetailModalProps) {
+  useEffect(() => {
+    function onKeyDown(event: KeyboardEvent) {
+      if (event.key === "Escape") onClose();
+    }
+    window.addEventListener("keydown", onKeyDown);
+    return () => window.removeEventListener("keydown", onKeyDown);
+  }, [onClose]);
+
+  const savedDate = new Date(tryItem.created_at).toLocaleDateString("ko-KR", {
+    year: "numeric",
+    month: "short",
+    day: "numeric",
+  });
+
+  return (
+    <div
+      role="dialog"
+      aria-modal="true"
+      aria-labelledby="try-detail-title"
+      className="fixed inset-0 z-[100] flex items-center justify-center bg-black/80 p-3 backdrop-blur-sm"
+      onClick={(event) => {
+        if (event.target === event.currentTarget) onClose();
+      }}
+    >
+      <div className="relative grid max-h-[92vh] w-full max-w-5xl overflow-hidden rounded-xl border border-white/10 bg-[#0a0a0a] text-white shadow-2xl md:grid-cols-[minmax(0,1.25fr)_380px]">
+        <button
+          type="button"
+          aria-label="Close"
+          onClick={onClose}
+          className="absolute right-3 top-3 z-10 rounded-full bg-white/10 p-2 text-white/80 hover:bg-white/20"
+        >
+          <X size={18} />
+        </button>
+
+        <div className="grid gap-3 overflow-y-auto p-4 md:grid-cols-2 md:p-5">
+          <figure className="md:col-span-2">
+            <div className="relative aspect-[3/4] overflow-hidden rounded-lg bg-white/5">
+              <Image
+                src={tryItem.image_url}
+                alt="Try-on result"
+                fill
+                className="object-contain"
+                sizes="(max-width: 768px) 94vw, 58vw"
+              />
+            </div>
+            <figcaption className="mt-2 text-xs uppercase tracking-wide text-white/50">
+              Result
+            </figcaption>
+          </figure>
+
+          {tryItem.person_original_image ? (
+            <figure>
+              <div className="relative aspect-[3/4] overflow-hidden rounded-lg bg-white/5">
+                <Image
+                  src={tryItem.person_original_image}
+                  alt="Uploaded original"
+                  fill
+                  className="object-contain"
+                  sizes="(max-width: 768px) 45vw, 240px"
+                />
+              </div>
+              <figcaption className="mt-2 text-xs uppercase tracking-wide text-white/50">
+                Original
+              </figcaption>
+            </figure>
+          ) : (
+            <div className="rounded-lg border border-white/10 p-4 text-sm text-white/60">
+              Original image unavailable
+            </div>
+          )}
+
+          {tryItem.source_post?.image_url ? (
+            <figure>
+              <div className="relative aspect-[3/4] overflow-hidden rounded-lg bg-white/5">
+                <Image
+                  src={tryItem.source_post.image_url}
+                  alt="Source post"
+                  fill
+                  className="object-cover"
+                  sizes="(max-width: 768px) 45vw, 240px"
+                />
+              </div>
+              <figcaption className="mt-2 text-xs uppercase tracking-wide text-white/50">
+                Source post
+              </figcaption>
+            </figure>
+          ) : (
+            <div className="rounded-lg border border-white/10 p-4 text-sm text-white/60">
+              Source post unavailable
+            </div>
+          )}
+        </div>
+
+        <aside className="overflow-y-auto border-t border-white/10 p-5 md:border-l md:border-t-0">
+          <p className="text-xs uppercase tracking-wide text-white/50">
+            Saved {savedDate}
+          </p>
+          <h2 id="try-detail-title" className="mt-2 text-xl font-semibold">
+            Try-on details
+          </h2>
+
+          <section className="mt-6">
+            <h3 className="text-sm font-medium text-white/70">Source post</h3>
+            {tryItem.source_post ? (
+              <div className="mt-3 space-y-2">
+                <p className="font-medium">
+                  {tryItem.source_post.title ?? "Untitled post"}
+                </p>
+                <p className="text-sm text-white/60">
+                  {tryItem.source_post.artist_name ??
+                    tryItem.source_post.group_name ??
+                    tryItem.source_post.context ??
+                    "Source context unavailable"}
+                </p>
+                {tryItem.source_post_id ? (
+                  <Link
+                    href={`/posts/${tryItem.source_post_id}`}
+                    className="inline-flex text-sm underline"
+                  >
+                    View post
+                  </Link>
+                ) : null}
+              </div>
+            ) : (
+              <p className="mt-3 text-sm text-white/60">
+                Source post unavailable
+              </p>
+            )}
+          </section>
+
+          <section className="mt-6">
+            <h3 className="text-sm font-medium text-white/70">
+              Selected items
+            </h3>
+            {tryItem.selected_items.length > 0 ? (
+              <div className="mt-3 space-y-3">
+                {tryItem.selected_items.map((item) => (
+                  <article
+                    key={item.id}
+                    className="grid grid-cols-[56px_minmax(0,1fr)] gap-3 rounded-lg border border-white/10 p-2"
+                  >
+                    <div className="relative h-14 w-14 overflow-hidden rounded bg-white/5">
+                      <Image
+                        src={item.thumbnail_url}
+                        alt={item.title}
+                        fill
+                        className="object-cover"
+                        sizes="56px"
+                      />
+                    </div>
+                    <div className="min-w-0">
+                      <p className="truncate text-sm font-medium">
+                        {item.title}
+                      </p>
+                      {item.description ? (
+                        <p className="line-clamp-2 text-xs text-white/60">
+                          {item.description}
+                        </p>
+                      ) : null}
+                      {item.keywords?.length ? (
+                        <div className="mt-1 flex flex-wrap gap-1">
+                          {item.keywords.map((keyword) => (
+                            <span
+                              key={keyword}
+                              className="rounded bg-white/10 px-1.5 py-0.5 text-[10px] text-white/70"
+                            >
+                              {keyword}
+                            </span>
+                          ))}
+                        </div>
+                      ) : null}
+                    </div>
+                  </article>
+                ))}
+              </div>
+            ) : (
+              <p className="mt-3 text-sm text-white/60">Items unavailable</p>
+            )}
+          </section>
+        </aside>
+      </div>
+    </div>
+  );
+}
+```
+
+- [ ] **Step 5: Verify modal tests**
+
+Run:
+
+```bash
+cd packages/web
+bun run test:unit lib/components/profile/__tests__/TryDetailModal.test.tsx
+```
+
+Expected: all modal tests pass.
+
+## Task 6: Wire Modal into TriesGrid
+
+**Files:**
+- Modify: `packages/web/lib/components/profile/TriesGrid.tsx`
+- Create: `packages/web/lib/components/profile/__tests__/TriesGrid.test.tsx`
+
+- [ ] **Step 1: Add failing grid interaction test**
+
+Create `TriesGrid.test.tsx`.
+
+```tsx
+/**
+ * @vitest-environment jsdom
+ */
+import React from "react";
+import { QueryClient, QueryClientProvider } from "@tanstack/react-query";
+import { describe, expect, it, vi } from "vitest";
+import { fireEvent, render, screen, waitFor } from "@testing-library/react";
+import "@testing-library/jest-dom";
+import { TriesGrid } from "../TriesGrid";
+
+vi.mock("next/image", () => ({
+  default: (props: React.ImgHTMLAttributes<HTMLImageElement>) => (
+    <img {...props} />
+  ),
+}));
+
+vi.mock("@/lib/api/generated/users/users", () => ({
+  getMyTries: vi.fn(async () => ({
+    data: [
+      {
+        id: "try-1",
+        image_url: "https://example.com/result.png",
+        created_at: "2026-05-07T00:00:00Z",
+        source_post_id: null,
+        selected_item_ids: [],
+        person_original_image: null,
+        source_post: null,
+        selected_items: [],
+      },
+    ],
+    pagination: {
+      current_page: 1,
+      per_page: 20,
+      total: 1,
+      total_pages: 1,
+    },
+  })),
+}));
+
+function renderGrid() {
+  const queryClient = new QueryClient({
+    defaultOptions: { queries: { retry: false } },
+  });
+  return render(
+    <QueryClientProvider client={queryClient}>
+      <TriesGrid />
+    </QueryClientProvider>
+  );
+}
+
+describe("TriesGrid", () => {
+  it("opens the detail modal when a try card is clicked", async () => {
+    renderGrid();
+
+    await waitFor(() =>
+      expect(screen.getByAltText("Try-on result")).toBeInTheDocument()
+    );
+    fireEvent.click(screen.getByRole("button", { name: /open try-on/i }));
+
+    expect(screen.getByRole("dialog")).toBeInTheDocument();
+    expect(screen.getByText("Try-on details")).toBeInTheDocument();
+  });
+});
+```
+
+- [ ] **Step 2: Run failing grid test**
+
+Run:
+
+```bash
+cd packages/web
+bun run test:unit lib/components/profile/__tests__/TriesGrid.test.tsx
+```
+
+Expected: the test fails because cards have no accessible name and no modal state.
+
+- [ ] **Step 3: Implement selected try state**
+
+In `TriesGrid.tsx`, import `useState` and `TryDetailModal`.
+
+```ts
+import { useRef, useEffect, useState } from "react";
+import { TryDetailModal } from "./TryDetailModal";
+```
+
+Inside `TriesGrid`, add:
+
+```ts
+const [selectedTry, setSelectedTry] = useState<TryItem | null>(null);
+```
+
+Update the card button.
+
+```tsx
+<button
+  key={t.id}
+  type="button"
+  aria-label="Open try-on details"
+  onClick={() => setSelectedTry(t)}
+  className="group relative overflow-hidden rounded-xl border border-border bg-card transition-all hover:border-primary/30"
+>
+```
+
+Wrap the returned grid in a fragment and render the modal after the grid.
+
+```tsx
+return (
+  <>
+    <div className={cn("grid grid-cols-2 md:grid-cols-3 gap-3", className)}>
+      {/* existing grid body */}
+    </div>
+    {selectedTry ? (
+      <TryDetailModal
+        tryItem={selectedTry}
+        onClose={() => setSelectedTry(null)}
+      />
+    ) : null}
+  </>
+);
+```
+
+- [ ] **Step 4: Verify grid test**
+
+Run:
+
+```bash
+cd packages/web
+bun run test:unit lib/components/profile/__tests__/TriesGrid.test.tsx
+```
+
+Expected: the grid click test passes.
+
+## Task 7: Full Verification and Manual QA
+
+**Files:**
+- Verify only.
+
+- [ ] **Step 1: Run targeted API and UI tests**
+
+Run:
+
+```bash
+cd packages/web
+bun run test:unit app/api/v1/tries/__tests__/route.test.ts app/api/v1/users/me/tries/__tests__/route.test.ts lib/components/profile/__tests__/TryDetailModal.test.tsx lib/components/profile/__tests__/TriesGrid.test.tsx
+```
+
+Expected: all targeted tests pass.
+
+- [ ] **Step 2: Run package typecheck**
+
+Run:
+
+```bash
+cd packages/web
+bun run typecheck
+```
+
+Expected: no TypeScript errors, especially around generated `TryItem`.
+
+- [ ] **Step 3: Run package lint**
+
+Run:
+
+```bash
+cd packages/web
+bun run lint
+```
+
+Expected: no lint errors in changed files.
+
+- [ ] **Step 4: Run manual QA on `/profile`**
+
+Start the app if it is not already running.
+
+```bash
+cd packages/web
+bun run dev
+```
+
+Manual checks:
+
+1. Save a new try-on from a VTON source post.
+2. Open `http://localhost:3000/profile`.
+3. Switch to the `Tries` tab.
+4. Click the saved result.
+5. Confirm the modal shows result image, uploaded original, source post image, source post context, selected items, and saved date.
+6. Click an old/legacy try-on row and confirm the modal opens with fallback text instead of breaking.
+7. Close via close button, backdrop click, and Escape.
+
+Expected: new and old tries both open in-place without profile navigation.
+
+## Self-Review
+
+- Spec coverage: Goals 1 and 2 are covered by Tasks 5 and 6. Goals 3 and 4 are covered by Tasks 1, 2, and 4. Goal 5 is covered by Task 2 fallback handling and Task 5 fallback UI.
+- Non-goals: no migration, no public history exposure, no retry action, and no broad VTON redesign are included.
+- Generated API models: Task 3 prevents manual edits under `packages/web/lib/api/generated/`.
+- Follow-ups: Supabase Storage, retry action, and delete action remain outside this plan.
diff --git a/docs/superpowers/plans/2026-05-14-content-studio-pipeline-v2.md b/docs/superpowers/plans/2026-05-14-content-studio-pipeline-v2.md
new file mode 100644
index 00000000..270ee328
--- /dev/null
+++ b/docs/superpowers/plans/2026-05-14-content-studio-pipeline-v2.md
@@ -0,0 +1,1892 @@
+# Content Studio Pipeline v2 Implementation Plan
+
+> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
+
+**Goal:** post 검색/선택 → 키워드 추출 → 마케팅 프롬프트 생성 → 3종 채널별 썸네일 자동 생성을 하나의 플로우로 완성하고, Research/Firecrawl 외부 의존성을 전체 제거한다.
+
+**Architecture:** 순차 스텝 파이프라인. PostPickerModal로 post 선택 → UUID 기반 Packet 생성 → 단일 OpenAI structured output 호출로 keywords + imagePrompts + variants 통합 생성 → gpt-image-2 edit 모드로 4종 썸네일 병렬 생성. Research 모듈 전체 삭제 및 관련 필드/참조 정리.
+
+**Tech Stack:** Next.js 16, React 19, Zod, OpenAI API (gpt-4.1 + gpt-image-2), Supabase Storage, Meilisearch (Orval generated `search()`), vitest
+
+---
+
+## File Structure
+
+### 새 파일
+
+| 파일                                                                        | 역할                 |
+| --------------------------------------------------------------------------- | -------------------- |
+| `packages/web/app/admin/content-studio/PostPickerModal.tsx`                 | Post 검색/선택 Modal |
+| `packages/web/app/api/v1/content/packets/[id]/generate-thumbnails/route.ts` | 썸네일 생성 API      |
+
+### 수정 파일
+
+| 파일                                                                      | 변경 내용                                                 |
+| ------------------------------------------------------------------------- | --------------------------------------------------------- |
+| `packages/web/lib/content-studio/packet-builder.ts`                       | UUID 생성 + 규칙 키워드 추출                              |
+| `packages/web/lib/content-studio/schemas.ts`                              | research 스키마/필드 삭제, unified 스키마 추가            |
+| `packages/web/lib/content-studio/llm-schemas.ts`                          | `unifiedContentResponseSchema` 추가                       |
+| `packages/web/lib/content-studio/llm-generation.ts`                       | `generateUnifiedContent()` 교체                           |
+| `packages/web/lib/content-studio/llm-prompts.ts`                          | 통합 프롬프트 교체                                        |
+| `packages/web/lib/content-studio/assets/openai-client.ts`                 | `generateChannelThumbnails()` 추가, 기본 모델 변경        |
+| `packages/web/lib/content-studio/governance-check.ts`                     | research 규칙 제거                                        |
+| `packages/web/lib/content-studio/index.ts`                                | research export 제거                                      |
+| `packages/web/app/admin/content-studio/page.tsx`                          | PostPickerModal 통합, research state 제거, 썸네일 UI 추가 |
+| `packages/web/app/admin/content-studio/AssetPanel.tsx`                    | research props 제거                                       |
+| `packages/web/app/admin/content-studio/ShortFormPanel.tsx`                | research props 제거                                       |
+| `packages/web/app/api/v1/content/packets/[id]/generate-variants/route.ts` | research 파라미터 제거                                    |
+| `packages/web/lib/content-studio/__tests__/llm-generation.test.ts`        | research 테스트 정리                                      |
+| `packages/web/lib/content-studio/__tests__/content-studio.test.ts`        | 통합 테스트 업데이트                                      |
+
+### 삭제 파일
+
+| 파일                                                                     | 이유                  |
+| ------------------------------------------------------------------------ | --------------------- |
+| `packages/web/lib/content-studio/research/` (전체)                       | Firecrawl 의존성 제거 |
+| `packages/web/lib/content-studio/__tests__/research.test.ts`             | 삭제 대상 테스트      |
+| `packages/web/app/admin/content-studio/ResearchPanel.tsx`                | 삭제 대상 UI          |
+| `packages/web/app/admin/content-studio/__tests__/ResearchPanel.test.tsx` | 삭제 대상 테스트      |
+| `packages/web/app/api/v1/content/research/route.ts`                      | 삭제 대상 API         |
+| `packages/web/app/api/v1/content/research/__tests__/route.test.ts`       | 삭제 대상 테스트      |
+
+---
+
+## Task 1: Packet ID UUID 수정
+
+**Files:**
+
+- Modify: `packages/web/lib/content-studio/packet-builder.ts`
+- Test: `packages/web/lib/content-studio/__tests__/content-studio.test.ts`
+
+- [ ] **Step 1: 기존 테스트 실행하여 baseline 확인**
+
+Run: `cd packages/web && npx vitest run lib/content-studio/__tests__/content-studio.test.ts`
+Expected: 모든 테스트 PASS
+
+- [ ] **Step 2: packet-builder.ts에서 packetId, variantId 함수 수정**
+
+`packages/web/lib/content-studio/packet-builder.ts`에서:
+
+```typescript
+// BEFORE (line ~230-240):
+function packetId(postId: string): string {
+  return `packet_${postId}`;
+}
+
+function variantId(packetIdValue: string, format: string): string {
+  return `${packetIdValue}_${format}`;
+}
+```
+
+→
+
+```typescript
+function packetId(): string {
+  return crypto.randomUUID();
+}
+
+function variantId(): string {
+  return crypto.randomUUID();
+}
+```
+
+`buildContentPacketFromPost` 함수 내에서 `packetId(post.id)` 호출을 `packetId()`로 변경.
+`generateChannelVariants` 함수 내에서 `variantId(packet.id, format)` 호출을 `variantId()`로 변경.
+
+- [ ] **Step 3: 테스트 업데이트**
+
+`packages/web/lib/content-studio/__tests__/content-studio.test.ts`에서 packet ID 형식 검증 부분을:
+
+- `packet_` prefix 대신 UUID 형식 검증으로 변경
+- `expect(packet.id).toMatch(/^[0-9a-f]{8}-/)` 패턴 사용
+
+- [ ] **Step 4: 테스트 실행**
+
+Run: `cd packages/web && npx vitest run lib/content-studio/__tests__/content-studio.test.ts`
+Expected: PASS
+
+- [ ] **Step 5: 커밋**
+
+```bash
+git add packages/web/lib/content-studio/packet-builder.ts packages/web/lib/content-studio/__tests__/content-studio.test.ts
+git commit -m "fix(content-studio): use crypto.randomUUID() for packet/variant IDs
+
+Replace packet_{postId} format that caused 'invalid input syntax for type uuid'
+errors. DB uniqueness is enforced by content_packets.post_id unique constraint."
+```
+
+---
+
+## Task 2: Research 스키마/필드 삭제 + Unified 스키마 추가
+
+**Files:**
+
+- Modify: `packages/web/lib/content-studio/schemas.ts`
+- Modify: `packages/web/lib/content-studio/llm-schemas.ts`
+
+- [ ] **Step 1: schemas.ts에서 research 관련 스키마 및 필드 삭제**
+
+`packages/web/lib/content-studio/schemas.ts`에서 삭제:
+
+1. 라인 28-38: `researchSourceTypeSchema`, `researchStatusSchema`, `researchConfidenceSchema`, `researchClaimTypeSchema`
+2. 라인 40-83: `researchSourceSchema`, `researchInsightSchema`, `researchRecommendationsSchema`, `researchRunSchema`
+3. 라인 121-127: `contentPacketSchema.externalEvidence` optional 필드
+4. 라인 148-165: `contentVariantSchema`의 `researchProvenance`, `claims`, `missingFacts` optional 필드
+5. 라인 185-193: `generateVariantsRequestSchema`의 `researchContext`, `useResearchInCopy` 필드
+6. 라인 195-200: `runResearchRequestSchema` 전체
+7. 라인 225: `assetPlanSchema`의 `researchRunId` 필드
+8. 라인 251: `assetPlanSchema.provenance`의 `sourceResearchRunId` 필드
+9. 라인 257: `shortFormPlanSchema`의 `researchRunId` 필드
+10. 라인 283: `shortFormPlanSchema.provenance`의 `sourceResearchRunId` 필드
+11. 라인 286-294: `assetPlanRequestSchema`의 `researchRun`, `useResearchInCopy` 필드
+12. 라인 296-304: `shortFormPlanRequestSchema`의 `researchRun`, `useResearchInCopy` 필드
+13. 라인 316-322: Research 관련 type export (`ResearchSourceType`, `ResearchSource`, `ResearchInsight`, `ResearchRecommendations`, `ResearchRun`, `RunResearchRequest`)
+
+수정 후 `schemas.ts`의 전체 내용은:
+
+```typescript
+import { z } from "zod";
+
+export const contentChannelSchema = z.enum(["instagram", "youtube", "x"]);
+export const contentVariantFormatSchema = z.enum([
+  "instagram_carousel",
+  "instagram_reel",
+  "youtube_shorts",
+  "x_thread",
+]);
+export const contentRiskLevelSchema = z.enum(["low", "medium", "high"]);
+export const contentReviewStatusSchema = z.enum([
+  "draft",
+  "needs_review",
+  "approved",
+  "rejected",
+]);
+export const governanceVerdictSchema = z.enum([
+  "approve",
+  "needs_review",
+  "reject",
+]);
+export const contentGenerationModeSchema = z.enum([
+  "template",
+  "hybrid",
+  "llm",
+]);
+
+export const itemEntitySchema = z.object({
+  id: z.string(),
+  title: z.string(),
+  brand: z.string().nullable(),
+  thumbnailUrl: z.string().nullable(),
+  sourceUrl: z.string().nullable(),
+  confidence: z.enum(["low", "medium", "high"]),
+  verified: z.boolean(),
+});
+
+export const contentPacketSchema = z.object({
+  id: z.string(),
+  postId: z.string(),
+  sourceImage: z.string(),
+  title: z.string(),
+  hook: z.string(),
+  artist: z.string().nullable(),
+  group: z.string().nullable(),
+  context: z.string().nullable(),
+  detectedItems: z.array(itemEntitySchema),
+  styleSummary: z.string(),
+  whyItWorks: z.string(),
+  alternatives: z.object({
+    budget: z.array(itemEntitySchema),
+    mid: z.array(itemEntitySchema),
+    premium: z.array(itemEntitySchema),
+  }),
+  disclosureFlags: z.object({
+    aiGenerated: z.boolean(),
+    syntheticMedia: z.boolean(),
+    sponsored: z.boolean(),
+    rightsRisk: z.boolean(),
+  }),
+  riskLevel: contentRiskLevelSchema,
+  reviewStatus: contentReviewStatusSchema,
+  createdAt: z.string(),
+});
+
+export const contentVariantSchema = z.object({
+  id: z.string(),
+  packetId: z.string(),
+  channel: contentChannelSchema,
+  format: contentVariantFormatSchema,
+  title: z.string(),
+  body: z.string(),
+  mediaPlan: z.record(z.string(), z.unknown()),
+  hashtags: z.array(z.string()),
+  disclosure: z.string(),
+  status: contentReviewStatusSchema,
+  generationMode: contentGenerationModeSchema.optional(),
+  llmModel: z.string().nullable().optional(),
+  promptVersion: z.string().nullable().optional(),
+  generationInputHash: z.string().nullable().optional(),
+  generationCost: z.record(z.string(), z.unknown()).nullable().optional(),
+  governanceResult: z.record(z.string(), z.unknown()).nullable().optional(),
+  riskNotes: z.array(z.string()).optional(),
+});
+
+export const governanceResultSchema = z.object({
+  verdict: governanceVerdictSchema,
+  riskLevel: contentRiskLevelSchema,
+  flags: z.array(z.string()),
+  requiredActions: z.array(z.string()),
+});
+
+export const createContentPacketRequestSchema = z.object({
+  postId: z.string().min(1),
+});
+
+export const generateVariantsRequestSchema = z.object({
+  packet: contentPacketSchema,
+  mode: contentGenerationModeSchema.default("template"),
+  channels: z.array(contentVariantFormatSchema).optional(),
+  locale: z.enum(["ko-KR", "en-US"]).default("ko-KR"),
+  tone: z.string().default("decoded_editorial"),
+});
+
+export const reviewVariantRequestSchema = z.object({
+  packet: contentPacketSchema,
+  variants: z.array(contentVariantSchema).default([]),
+});
+
+export const assetTargetFormatSchema = z.enum([
+  "instagram_feed",
+  "instagram_story",
+  "instagram_carousel_cover",
+  "youtube_thumbnail",
+  "pinterest_pin",
+]);
+
+export const assetEditModeSchema = z.enum(["generate", "edit", "variant"]);
+export const assetPlanStatusSchema = z.enum(["draft", "ready", "failed"]);
+export const shortFormPlatformSchema = z.enum([
+  "instagram_reel",
+  "youtube_shorts",
+]);
+
+export const assetPlanSchema = z.object({
+  id: z.string(),
+  packetId: z.string(),
+  status: assetPlanStatusSchema,
+  createdAt: z.string(),
+  updatedAt: z.string(),
+  imageAssets: z.array(
+    z.object({
+      id: z.string(),
+      format: assetTargetFormatSchema,
+      prompt: z.string(),
+      size: z.string(),
+      editMode: assetEditModeSchema,
+      previewUrl: z.string().url().nullable(),
+      altText: z.string(),
+    }),
+  ),
+  overlayText: z.array(
+    z.object({
+      id: z.string(),
+      format: assetTargetFormatSchema,
+      headline: z.string(),
+      subheadline: z.string().nullable(),
+    }),
+  ),
+  provenance: z.object({
+    sourcePacketId: z.string(),
+  }),
+});
+
+export const shortFormPlanSchema = z.object({
+  id: z.string(),
+  packetId: z.string(),
+  platform: shortFormPlatformSchema,
+  durationSeconds: z.number().int().min(6).max(60),
+  status: assetPlanStatusSchema,
+  createdAt: z.string(),
+  updatedAt: z.string(),
+  hook: z.string(),
+  voiceover: z.object({
+    text: z.string(),
+    tone: z.string(),
+    targetReadingSeconds: z.number().int().min(1).max(60),
+  }),
+  scenes: z.array(
+    z.object({
+      id: z.string(),
+      order: z.number().int().min(1),
+      seconds: z.number().int().min(1).max(30),
+      visualDirection: z.string(),
+      onScreenText: z.string(),
+      narration: z.string(),
+    }),
+  ),
+  cta: z.string(),
+  provenance: z.object({
+    sourcePacketId: z.string(),
+  }),
+});
+
+export const assetPlanRequestSchema = z.object({
+  packet: contentPacketSchema,
+  variants: z.array(contentVariantSchema).default([]),
+  assetTypes: z.array(assetTargetFormatSchema).min(1).max(5),
+  embedHeadline: z.boolean().optional(),
+  model: z.string().optional(),
+});
+
+export const shortFormPlanRequestSchema = z.object({
+  packet: contentPacketSchema,
+  variants: z.array(contentVariantSchema).default([]),
+  platform: shortFormPlatformSchema,
+  durationSeconds: z.number().int().min(6).max(60).default(20),
+  model: z.string().optional(),
+});
+
+export type ContentChannel = z.infer<typeof contentChannelSchema>;
+export type ContentVariantFormat = z.infer<typeof contentVariantFormatSchema>;
+export type ContentRiskLevel = z.infer<typeof contentRiskLevelSchema>;
+export type ContentReviewStatus = z.infer<typeof contentReviewStatusSchema>;
+export type ContentGenerationMode = z.infer<typeof contentGenerationModeSchema>;
+export type GovernanceVerdict = z.infer<typeof governanceVerdictSchema>;
+export type ItemEntity = z.infer<typeof itemEntitySchema>;
+export type ContentPacket = z.infer<typeof contentPacketSchema>;
+export type ContentVariant = z.infer<typeof contentVariantSchema>;
+export type GovernanceResult = z.infer<typeof governanceResultSchema>;
+export type AssetTargetFormat = z.infer<typeof assetTargetFormatSchema>;
+export type AssetEditMode = z.infer<typeof assetEditModeSchema>;
+export type AssetPlanStatus = z.infer<typeof assetPlanStatusSchema>;
+export type ShortFormPlatform = z.infer<typeof shortFormPlatformSchema>;
+export type AssetPlan = z.infer<typeof assetPlanSchema>;
+export type ShortFormPlan = z.infer<typeof shortFormPlanSchema>;
+export type AssetPlanRequest = z.infer<typeof assetPlanRequestSchema>;
+export type ShortFormPlanRequest = z.infer<typeof shortFormPlanRequestSchema>;
+```
+
+- [ ] **Step 2: llm-schemas.ts에 unifiedContentResponseSchema 추가**
+
+`packages/web/lib/content-studio/llm-schemas.ts` 끝에 추가:
+
+```typescript
+export const unifiedContentResponseSchema = z.object({
+  keywords: z.array(z.string()),
+  imagePrompts: z.object({
+    youtube: z.string(),
+    instagram_feed: z.string(),
+    instagram_story: z.string(),
+  }),
+  variants: z.array(
+    z.object({
+      channel: contentChannelSchema,
+      format: contentVariantFormatSchema,
+      title: z.string(),
+      body: z.string(),
+      hashtags: z.array(z.string()),
+    }),
+  ),
+});
+
+export type UnifiedContentResponse = z.infer<
+  typeof unifiedContentResponseSchema
+>;
+
+export const unifiedContentJsonSchema = {
+  name: "unified_content",
+  strict: true,
+  schema: {
+    type: "object",
+    required: ["keywords", "imagePrompts", "variants"],
+    additionalProperties: false,
+    properties: {
+      keywords: { type: "array", items: { type: "string" } },
+      imagePrompts: {
+        type: "object",
+        required: ["youtube", "instagram_feed", "instagram_story"],
+        additionalProperties: false,
+        properties: {
+          youtube: { type: "string" },
+          instagram_feed: { type: "string" },
+          instagram_story: { type: "string" },
+        },
+      },
+      variants: {
+        type: "array",
+        items: {
+          type: "object",
+          required: ["channel", "format", "title", "body", "hashtags"],
+          additionalProperties: false,
+          properties: {
+            channel: { type: "string", enum: ["instagram", "youtube", "x"] },
+            format: {
+              type: "string",
+              enum: [
+                "instagram_carousel",
+                "instagram_reel",
+                "youtube_shorts",
+                "x_thread",
+              ],
+            },
+            title: { type: "string" },
+            body: { type: "string" },
+            hashtags: { type: "array", items: { type: "string" } },
+          },
+        },
+      },
+    },
+  },
+} as const;
+```
+
+`llm-schemas.ts`에서 `llmContentVariantSchema`의 `claims` 필드에서 `source: z.enum(["packet", "research", "generated"])` → `source: z.enum(["packet", "generated"])`로 변경. `"research"` 옵션 제거.
+
+- [ ] **Step 3: 타입 체크 실행**
+
+Run: `cd packages/web && npx tsc --noEmit --pretty 2>&1 | head -50`
+Expected: research 관련 import 에러 다수 발생 (이는 예상된 것 — 다음 Task에서 해결)
+
+- [ ] **Step 4: 커밋**
+
+```bash
+git add packages/web/lib/content-studio/schemas.ts packages/web/lib/content-studio/llm-schemas.ts
+git commit -m "refactor(content-studio): remove research schemas, add unified content schema
+
+Drop all Research/Firecrawl types (ResearchRun, ResearchSource, ResearchInsight,
+etc.) and research-related fields from ContentPacket, ContentVariant,
+generateVariantsRequestSchema, assetPlanRequestSchema, shortFormPlanRequestSchema.
+
+Add unifiedContentResponseSchema for the new single-call LLM pipeline that
+generates keywords + imagePrompts + channel variants together."
+```
+
+---
+
+## Task 3: Research 파일 삭제 + 의존 코드 정리
+
+**Files:**
+
+- Delete: `packages/web/lib/content-studio/research/` (전체 디렉토리)
+- Delete: `packages/web/lib/content-studio/__tests__/research.test.ts`
+- Delete: `packages/web/app/admin/content-studio/ResearchPanel.tsx`
+- Delete: `packages/web/app/admin/content-studio/__tests__/ResearchPanel.test.tsx` (있는 경우)
+- Delete: `packages/web/app/api/v1/content/research/route.ts`
+- Delete: `packages/web/app/api/v1/content/research/__tests__/route.test.ts` (있는 경우)
+- Modify: `packages/web/lib/content-studio/index.ts`
+- Modify: `packages/web/lib/content-studio/governance-check.ts`
+- Modify: `packages/web/lib/content-studio/llm-generation.ts`
+- Modify: `packages/web/lib/content-studio/assets/openai-client.ts`
+- Modify: `packages/web/app/admin/content-studio/AssetPanel.tsx`
+- Modify: `packages/web/app/admin/content-studio/ShortFormPanel.tsx`
+- Modify: `packages/web/app/api/v1/content/packets/[id]/generate-variants/route.ts`
+
+- [ ] **Step 1: research 디렉토리 및 관련 파일 삭제**
+
+```bash
+rm -rf packages/web/lib/content-studio/research/
+rm -f packages/web/lib/content-studio/__tests__/research.test.ts
+rm -f packages/web/app/admin/content-studio/ResearchPanel.tsx
+rm -f packages/web/app/admin/content-studio/__tests__/ResearchPanel.test.tsx
+rm -rf packages/web/app/api/v1/content/research/
+```
+
+- [ ] **Step 2: index.ts에서 research export 제거**
+
+`packages/web/lib/content-studio/index.ts`에서:
+
+```typescript
+// BEFORE:
+export * from "./schemas";
+export * from "./packet-builder";
+export * from "./governance-check";
+export * from "./opportunity-score";
+export * from "./research/query-suggestions";
+export * from "./assets";
+```
+
+→
+
+```typescript
+export * from "./schemas";
+export * from "./packet-builder";
+export * from "./governance-check";
+export * from "./opportunity-score";
+export * from "./assets";
+```
+
+- [ ] **Step 3: governance-check.ts에서 research 규칙 제거**
+
+`packages/web/lib/content-studio/governance-check.ts`에서 research 관련 코드(라인 47-93) 전체 삭제. 수정 후:
+
+```typescript
+import type {
+  ContentPacket,
+  ContentVariant,
+  GovernanceResult,
+} from "./schemas";
+
+export function runGovernanceCheck(
+  packet: ContentPacket,
+  variants: ContentVariant[],
+): GovernanceResult {
+  const flags: string[] = [];
+  const requiredActions: string[] = [];
+
+  const hasInterpretation =
+    packet.detectedItems.length > 0 ||
+    (packet.styleSummary.trim().length > 0 &&
+      packet.styleSummary !== "No interpretation layer is available yet.");
+
+  if (!hasInterpretation) {
+    flags.push("thin_repost");
+    requiredActions.push("Add interpretation layer before publishing");
+  }
+
+  if (packet.disclosureFlags.rightsRisk) {
+    flags.push("celebrity_likeness_review");
+    requiredActions.push("Human review before publishing");
+  }
+
+  if (packet.disclosureFlags.syntheticMedia) {
+    flags.push("synthetic_media_disclosure");
+    requiredActions.push("Add synthetic media disclosure");
+  }
+
+  const lowConfidenceCount = packet.detectedItems.filter(
+    (item) => item.confidence === "low",
+  ).length;
+  if (lowConfidenceCount > 0) {
+    flags.push("low_confidence_item");
+    requiredActions.push("Verify low-confidence item identity");
+  }
+
+  if (variants.some((variant) => variant.disclosure.trim().length === 0)) {
+    flags.push("missing_disclosure");
+    requiredActions.push("Add disclosure copy");
+  }
+
+  if (flags.includes("thin_repost")) {
+    return {
+      verdict: "reject",
+      riskLevel: "high",
+      flags,
+      requiredActions,
+    };
+  }
+
+  if (packet.riskLevel !== "low" || flags.length > 0) {
+    return {
+      verdict: "needs_review",
+      riskLevel: packet.riskLevel,
+      flags,
+      requiredActions,
+    };
+  }
+
+  return {
+    verdict: "approve",
+    riskLevel: "low",
+    flags,
+    requiredActions,
+  };
+}
+```
+
+- [ ] **Step 4: llm-generation.ts에서 research 관련 로직 제거**
+
+`packages/web/lib/content-studio/llm-generation.ts`에서:
+
+1. import에서 `ResearchInsight`, `ResearchSource` 제거
+2. `LlmGenerate` 타입에서 `researchContext` 제거
+3. `mapLlmResponseToVariants`에서 `researchContext`, `useResearchInCopy` 매개변수 및 `researchProvenance` 매핑 제거
+4. `generateVariantsWithMode`에서 `researchContext`, `useResearchInCopy` 매개변수 제거
+5. `evidenceBackedInsights` 함수 삭제
+6. `researchContextForCopy` 변수 및 관련 hash 로직 제거
+
+수정 후 `generateVariantsWithMode` 시그니처:
+
+```typescript
+export async function generateVariantsWithMode(input: {
+  packet: ContentPacket;
+  mode: ContentGenerationMode;
+  channels?: ContentVariantFormat[];
+  locale?: "ko-KR" | "en-US";
+  tone?: string;
+  llmEnabled?: boolean;
+  model?: string;
+  llmGenerate?: LlmGenerate;
+}): Promise<GenerateVariantsWithModeResult>;
+```
+
+수정 후 `mapLlmResponseToVariants` 시그니처:
+
+```typescript
+function mapLlmResponseToVariants(input: {
+  packet: ContentPacket;
+  response: ContentVariantLLMResponse;
+  templates: ContentVariant[];
+  mode: ContentGenerationMode;
+  model: string;
+  hash: string;
+}): ContentVariant[];
+```
+
+return 문에서 `claims`, `researchProvenance` 필드 제거.
+
+수정 후 `LlmGenerate` 타입:
+
+```typescript
+type LlmGenerate = (input: {
+  packet: ContentPacket;
+  channels: ContentVariantFormat[];
+  locale: "ko-KR" | "en-US";
+  tone: string;
+  model?: string;
+}) => Promise<unknown>;
+```
+
+hash 계산에서 `researchContext` 제거:
+
+```typescript
+const hash = inputHash({
+  packet: input.packet,
+  channels,
+  locale,
+  tone,
+  promptVersion: CONTENT_STUDIO_PROMPT_VERSION,
+});
+```
+
+- [ ] **Step 5: openai-client.ts에서 ResearchRun import 제거**
+
+`packages/web/lib/content-studio/assets/openai-client.ts` 라인 1-9의 import에서 `ResearchRun` 제거. 파일 내에서 `ResearchRun` 타입을 사용하는 함수 시그니처가 있다면 해당 파라미터도 제거.
+
+- [ ] **Step 6: AssetPanel.tsx에서 research props 제거**
+
+`packages/web/app/admin/content-studio/AssetPanel.tsx`에서:
+
+1. `ResearchRun` import 제거
+2. Props에서 `researchRun: ResearchRun | null` 및 `useResearchInCopy: boolean` 제거
+3. 함수 내 `researchRun`, `useResearchInCopy` destructuring 및 사용 제거
+4. API 호출 body에서 `researchRun`, `useResearchInCopy` 제거
+
+- [ ] **Step 7: ShortFormPanel.tsx에서 research props 제거**
+
+`packages/web/app/admin/content-studio/ShortFormPanel.tsx`에서:
+
+1. `ResearchRun` import 제거
+2. Props에서 `researchRun: ResearchRun | null` 및 `useResearchInCopy: boolean` 제거
+3. 함수 내 `researchRun`, `useResearchInCopy` destructuring 및 사용 제거
+4. API 호출 body에서 `researchRun`, `useResearchInCopy` 제거
+
+- [ ] **Step 8: generate-variants route.ts에서 research 파라미터 제거**
+
+`packages/web/app/api/v1/content/packets/[id]/generate-variants/route.ts`에서:
+
+1. `generateVariantsWithMode` 호출 시 `researchContext`, `useResearchInCopy` 전달 제거
+2. `generateVariantsRequestSchema` 파싱 후 해당 필드 참조 제거
+
+- [ ] **Step 9: llm-generation.test.ts 정리**
+
+`packages/web/lib/content-studio/__tests__/llm-generation.test.ts`에서:
+
+1. research context 관련 테스트 케이스 삭제 (researchContext 파라미터를 사용하는 모든 it/test 블록)
+2. mock data에서 `researchProvenance`, `claims`, `missingFacts` 필드 제거
+3. research import 제거
+
+- [ ] **Step 10: 전체 테스트 실행**
+
+Run: `cd packages/web && npx vitest run lib/content-studio/__tests__/ 2>&1 | tail -30`
+Expected: PASS (research.test.ts는 이미 삭제됨)
+
+- [ ] **Step 11: 타입 체크**
+
+Run: `cd packages/web && npx tsc --noEmit --pretty 2>&1 | head -30`
+Expected: 0 errors
+
+- [ ] **Step 12: 커밋**
+
+```bash
+git add -A packages/web/lib/content-studio/ packages/web/app/admin/content-studio/ packages/web/app/api/v1/content/research/ packages/web/app/api/v1/content/packets/
+git commit -m "refactor(content-studio): remove Research/Firecrawl module entirely
+
+Delete research/ directory (7 files), ResearchPanel, research API route,
+research tests. Clean up governance-check (remove 4 research-only rules),
+llm-generation (remove researchContext/useResearchInCopy params),
+AssetPanel/ShortFormPanel (remove research props), and generate-variants
+route (remove research params from request body)."
+```
+
+---
+
+## Task 4: 규칙 키워드 추출 + 통합 LLM 생성
+
+**Files:**
+
+- Modify: `packages/web/lib/content-studio/packet-builder.ts`
+- Modify: `packages/web/lib/content-studio/llm-generation.ts`
+- Modify: `packages/web/lib/content-studio/llm-prompts.ts`
+- Test: `packages/web/lib/content-studio/__tests__/llm-generation.test.ts`
+
+- [ ] **Step 1: packet-builder.ts에 규칙 키워드 추출 함수 추가**
+
+`packages/web/lib/content-studio/packet-builder.ts`에 추가:
+
+```typescript
+export function extractRuleKeywords(packet: ContentPacket): string[] {
+  const keywords: string[] = [];
+  if (packet.artist) keywords.push(packet.artist);
+  if (packet.group) keywords.push(packet.group);
+  for (const item of packet.detectedItems) {
+    if (item.brand) keywords.push(item.brand);
+    keywords.push(item.title);
+  }
+  if (packet.context) keywords.push(packet.context);
+  return [...new Set(keywords.filter(Boolean))];
+}
+```
+
+- [ ] **Step 2: 키워드 추출 테스트 작성**
+
+`packages/web/lib/content-studio/__tests__/content-studio.test.ts`에 추가:
+
+```typescript
+import { extractRuleKeywords } from "../packet-builder";
+
+describe("extractRuleKeywords", () => {
+  it("extracts artist, group, brand, item titles, and context", () => {
+    const packet = {
+      ...minimalPacket,
+      artist: "IU",
+      group: "solo",
+      context: "K-pop fashion",
+      detectedItems: [
+        {
+          id: "1",
+          title: "White Jacket",
+          brand: "Miu Miu",
+          thumbnailUrl: null,
+          sourceUrl: null,
+          confidence: "high" as const,
+          verified: false,
+        },
+        {
+          id: "2",
+          title: "Black Boots",
+          brand: null,
+          thumbnailUrl: null,
+          sourceUrl: null,
+          confidence: "medium" as const,
+          verified: false,
+        },
+      ],
+    };
+    const keywords = extractRuleKeywords(packet);
+    expect(keywords).toContain("IU");
+    expect(keywords).toContain("solo");
+    expect(keywords).toContain("Miu Miu");
+    expect(keywords).toContain("White Jacket");
+    expect(keywords).toContain("Black Boots");
+    expect(keywords).toContain("K-pop fashion");
+    expect(new Set(keywords).size).toBe(keywords.length);
+  });
+});
+```
+
+(`minimalPacket`는 기존 테스트 파일에 이미 정의된 fixture를 사용. 없다면 적절한 ContentPacket mock 생성.)
+
+- [ ] **Step 3: 테스트 실행**
+
+Run: `cd packages/web && npx vitest run lib/content-studio/__tests__/content-studio.test.ts`
+Expected: PASS
+
+- [ ] **Step 4: llm-prompts.ts에 통합 프롬프트 작성**
+
+`packages/web/lib/content-studio/llm-prompts.ts` 전체 교체:
+
+```typescript
+export const CONTENT_STUDIO_PROMPT_VERSION = "content-studio-v1.0.0";
+
+export function buildUnifiedContentPrompt(
+  packet: {
+    title: string;
+    artist: string | null;
+    group: string | null;
+    context: string | null;
+    styleSummary: string;
+    whyItWorks: string;
+  },
+  ruleKeywords: string[],
+): string {
+  const lines = [
+    "You are an editorial content engine for decoded, a fashion/style discovery platform.",
+    "",
+    `Post title: ${packet.title}`,
+    packet.artist ? `Artist: ${packet.artist}` : null,
+    packet.group ? `Group: ${packet.group}` : null,
+    packet.context ? `Context: ${packet.context}` : null,
+    `Style summary: ${packet.styleSummary}`,
+    `Why it works: ${packet.whyItWorks}`,
+    "",
+    `Rule-based keywords (must include): ${ruleKeywords.join(", ")}`,
+    "",
+    "Generate:",
+    "1. keywords: include all rule-based keywords plus additional marketing keywords (max 15 total)",
+    "2. imagePrompts: creative prompts for thumbnail generation per channel. Reference the post's visual style.",
+    "   - youtube: 1536×1024 landscape, bold text overlay friendly",
+    "   - instagram_feed: 1024×1024 square, clean composition",
+    "   - instagram_story: 1024×1536 vertical, mobile-first",
+    "3. variants: marketing copy for each channel (youtube, instagram, x)",
+    "   - youtube: { channel: 'youtube', format: 'youtube_shorts', title, body, hashtags }",
+    "   - instagram: { channel: 'instagram', format: 'instagram_carousel', title, body, hashtags }",
+    "   - x: { channel: 'x', format: 'x_thread', title, body, hashtags }",
+    "",
+    "Rules:",
+    "- Write in Korean (ko-KR) by default",
+    "- No claims without confidence",
+    "- Include interpretation (annotation, explanation, comparison)",
+    "- Keep hashtags relevant and concise (5-10 per channel)",
+    "- Image prompts should describe visual composition, not just text",
+  ].filter((line): line is string => line !== null);
+
+  return lines.join("\n");
+}
+```
+
+- [ ] **Step 5: llm-generation.ts에 generateUnifiedContent 추가**
+
+`packages/web/lib/content-studio/llm-generation.ts`에 새 함수 추가:
+
+```typescript
+import {
+  unifiedContentResponseSchema,
+  type UnifiedContentResponse,
+} from "./llm-schemas";
+import { buildUnifiedContentPrompt } from "./llm-prompts";
+import { extractRuleKeywords } from "./packet-builder";
+
+export type UnifiedContentResult = {
+  keywords: string[];
+  imagePrompts: {
+    youtube: string;
+    instagram_feed: string;
+    instagram_story: string;
+  };
+  variants: ContentVariant[];
+  mode: ContentGenerationMode;
+  warning: string | null;
+};
+
+export async function generateUnifiedContent(input: {
+  packet: ContentPacket;
+  mode?: ContentGenerationMode;
+  locale?: "ko-KR" | "en-US";
+  llmEnabled?: boolean;
+  model?: string;
+}): Promise<UnifiedContentResult> {
+  const mode = input.mode ?? "llm";
+  const ruleKeywords = extractRuleKeywords(input.packet);
+  const templates = generateChannelVariants(input.packet);
+  const hash = inputHash({
+    packet: input.packet,
+    promptVersion: CONTENT_STUDIO_PROMPT_VERSION,
+    ruleKeywords,
+  });
+
+  const emptyResult: UnifiedContentResult = {
+    keywords: ruleKeywords,
+    imagePrompts: { youtube: "", instagram_feed: "", instagram_story: "" },
+    variants: withGenerationMetadata(templates, {
+      mode: "template",
+      model: null,
+      hash,
+    }),
+    mode: "template",
+    warning: null,
+  };
+
+  if (mode === "template") {
+    return emptyResult;
+  }
+
+  const llmEnabled = input.llmEnabled ?? isContentStudioLlmEnabled();
+  const model = input.model ?? contentStudioModel();
+
+  if (!llmEnabled || !process.env.OPENAI_API_KEY) {
+    return {
+      ...emptyResult,
+      warning: "LLM client is not configured; used template fallback.",
+    };
+  }
+
+  try {
+    const prompt = buildUnifiedContentPrompt(input.packet, ruleKeywords);
+    const raw = await generateContentVariantsWithOpenAI({
+      packet: input.packet,
+      channels: ["instagram_carousel", "youtube_shorts", "x_thread"],
+      locale: input.locale ?? "ko-KR",
+      tone: "decoded_editorial",
+      model,
+    });
+    const parsed = unifiedContentResponseSchema.safeParse(raw);
+
+    if (!parsed.success) {
+      return {
+        ...emptyResult,
+        warning: "LLM output failed validation; used template fallback.",
+      };
+    }
+
+    const merged = templates.map((template) => {
+      const generated = parsed.data.variants.find(
+        (v) => v.format === template.format,
+      );
+      if (!generated) return template;
+      return {
+        ...template,
+        title: generated.title,
+        body: generated.body,
+        hashtags: generated.hashtags,
+        generationMode: mode as ContentGenerationMode,
+        llmModel: model,
+        promptVersion: CONTENT_STUDIO_PROMPT_VERSION,
+        generationInputHash: hash,
+      };
+    });
+
+    return {
+      keywords: parsed.data.keywords,
+      imagePrompts: parsed.data.imagePrompts,
+      variants: merged,
+      mode,
+      warning: null,
+    };
+  } catch (error) {
+    return {
+      ...emptyResult,
+      warning:
+        error instanceof Error
+          ? `LLM generation failed: ${error.message}; used template fallback.`
+          : "LLM generation failed; used template fallback.",
+    };
+  }
+}
+```
+
+**주의**: `generateContentVariantsWithOpenAI` 함수의 실제 구현을 확인하고, unified JSON schema를 사용하도록 호출부를 조정해야 합니다. `llm-client.ts`의 OpenAI 호출에서 `response_format`의 JSON schema를 `unifiedContentJsonSchema`로 교체해야 할 수 있습니다. 실제 구현 시 `llm-client.ts`를 읽고 적절히 수정하세요.
+
+- [ ] **Step 6: 통합 LLM 테스트 작성**
+
+`packages/web/lib/content-studio/__tests__/llm-generation.test.ts`에 추가:
+
+```typescript
+import { generateUnifiedContent } from "../llm-generation";
+
+describe("generateUnifiedContent", () => {
+  it("returns template fallback when LLM disabled", async () => {
+    const result = await generateUnifiedContent({
+      packet: mockPacket,
+      mode: "llm",
+      llmEnabled: false,
+    });
+
+    expect(result.mode).toBe("template");
+    expect(result.warning).toContain("template fallback");
+    expect(result.keywords.length).toBeGreaterThan(0);
+    expect(result.variants.length).toBe(4);
+  });
+
+  it("returns rule keywords even in template mode", async () => {
+    const packet = {
+      ...mockPacket,
+      artist: "BLACKPINK",
+      group: "BLACKPINK",
+    };
+    const result = await generateUnifiedContent({
+      packet,
+      mode: "template",
+    });
+
+    expect(result.keywords).toContain("BLACKPINK");
+  });
+});
+```
+
+(`mockPacket`는 기존 테스트 파일의 mock 데이터를 재사용.)
+
+- [ ] **Step 7: 테스트 실행**
+
+Run: `cd packages/web && npx vitest run lib/content-studio/__tests__/llm-generation.test.ts`
+Expected: PASS
+
+- [ ] **Step 8: 커밋**
+
+```bash
+git add packages/web/lib/content-studio/packet-builder.ts packages/web/lib/content-studio/llm-generation.ts packages/web/lib/content-studio/llm-prompts.ts packages/web/lib/content-studio/__tests__/
+git commit -m "feat(content-studio): add unified LLM content generation
+
+Single OpenAI call generates keywords, imagePrompts (3 channels), and
+marketing variants together. extractRuleKeywords() collects brand/artist/
+group/context from packet for mandatory inclusion. Template fallback
+preserved when LLM is disabled."
+```
+
+---
+
+## Task 5: 채널별 썸네일 생성 API
+
+**Files:**
+
+- Modify: `packages/web/lib/content-studio/assets/openai-client.ts`
+- Create: `packages/web/app/api/v1/content/packets/[id]/generate-thumbnails/route.ts`
+
+- [ ] **Step 1: openai-client.ts에 generateChannelThumbnails 함수 추가**
+
+`packages/web/lib/content-studio/assets/openai-client.ts`에 추가:
+
+```typescript
+export type ChannelThumbnailResult = {
+  thumbnails: Record<string, string>;
+  failures: Array<{ channel: string; error: string }>;
+};
+
+const THUMBNAIL_SIZES: Record<string, string> = {
+  youtube: "1536x1024",
+  x: "1536x1024",
+  instagram_feed: "1024x1024",
+  instagram_story: "1024x1536",
+};
+
+export async function generateChannelThumbnails(
+  packetId: string,
+  imagePrompts: {
+    youtube: string;
+    instagram_feed: string;
+    instagram_story: string;
+  },
+  referenceImageUrl?: string,
+): Promise<ChannelThumbnailResult> {
+  const channels = [
+    { key: "youtube", prompt: imagePrompts.youtube },
+    { key: "x", prompt: imagePrompts.youtube },
+    { key: "instagram_feed", prompt: imagePrompts.instagram_feed },
+    { key: "instagram_story", prompt: imagePrompts.instagram_story },
+  ];
+
+  const results = await Promise.allSettled(
+    channels.map(async ({ key, prompt }) => {
+      const size = THUMBNAIL_SIZES[key];
+      let imageBytes: { bytes: Buffer; contentType: string };
+
+      if (referenceImageUrl) {
+        try {
+          imageBytes = await fetchImageBytesFromOpenAIEdit(prompt, size, [
+            referenceImageUrl,
+          ]);
+        } catch {
+          imageBytes = await fetchImageBytesFromOpenAI(prompt, size);
+        }
+      } else {
+        imageBytes = await fetchImageBytesFromOpenAI(prompt, size);
+      }
+
+      const path = `content-studio/${packetId}/${key}.png`;
+      const url = await uploadToStorage(
+        imageBytes.bytes,
+        path,
+        imageBytes.contentType,
+      );
+      return { key, url };
+    }),
+  );
+
+  const thumbnails: Record<string, string> = {};
+  const failures: Array<{ channel: string; error: string }> = [];
+
+  for (const [i, result] of results.entries()) {
+    if (result.status === "fulfilled") {
+      thumbnails[result.value.key] = result.value.url;
+    } else {
+      failures.push({
+        channel: channels[i].key,
+        error:
+          result.reason instanceof Error
+            ? result.reason.message
+            : "Unknown error",
+      });
+    }
+  }
+
+  return { thumbnails, failures };
+}
+```
+
+- [ ] **Step 2: openai-client.ts 기본 모델 변경**
+
+`packages/web/lib/content-studio/assets/openai-client.ts`에서:
+
+```typescript
+// BEFORE:
+export function contentStudioImageModel(): string {
+  return process.env.CONTENT_STUDIO_IMAGE_MODEL || "gpt-image-1";
+}
+```
+
+→
+
+```typescript
+export function contentStudioImageModel(): string {
+  return process.env.CONTENT_STUDIO_IMAGE_MODEL || "gpt-image-2";
+}
+```
+
+- [ ] **Step 3: 썸네일 생성 API route 작성**
+
+`packages/web/app/api/v1/content/packets/[id]/generate-thumbnails/route.ts`:
+
+```typescript
+import { NextRequest, NextResponse } from "next/server";
+import { z } from "zod";
+import { generateChannelThumbnails } from "@/lib/content-studio/assets/openai-client";
+
+const generateThumbnailsRequestSchema = z.object({
+  imagePrompts: z.object({
+    youtube: z.string().min(1),
+    instagram_feed: z.string().min(1),
+    instagram_story: z.string().min(1),
+  }),
+  referenceImageUrl: z.string().url().optional(),
+});
+
+export async function POST(
+  request: NextRequest,
+  { params }: { params: Promise<{ id: string }> },
+) {
+  const { id: packetId } = await params;
+
+  const body = await request.json().catch(() => null);
+  if (!body) {
+    return NextResponse.json({ error: "Invalid JSON body" }, { status: 400 });
+  }
+
+  const parsed = generateThumbnailsRequestSchema.safeParse(body);
+  if (!parsed.success) {
+    return NextResponse.json(
+      { error: "Validation failed", details: parsed.error.flatten() },
+      { status: 400 },
+    );
+  }
+
+  try {
+    const result = await generateChannelThumbnails(
+      packetId,
+      parsed.data.imagePrompts,
+      parsed.data.referenceImageUrl,
+    );
+    return NextResponse.json(result);
+  } catch (error) {
+    return NextResponse.json(
+      {
+        error:
+          error instanceof Error
+            ? error.message
+            : "Thumbnail generation failed",
+      },
+      { status: 500 },
+    );
+  }
+}
+```
+
+- [ ] **Step 4: 타입 체크**
+
+Run: `cd packages/web && npx tsc --noEmit --pretty 2>&1 | head -20`
+Expected: 0 errors
+
+- [ ] **Step 5: 커밋**
+
+```bash
+git add packages/web/lib/content-studio/assets/openai-client.ts packages/web/app/api/v1/content/packets/\[id\]/generate-thumbnails/
+git commit -m "feat(content-studio): add channel thumbnail generation API
+
+POST /api/v1/content/packets/{id}/generate-thumbnails accepts imagePrompts
+(youtube, instagram_feed, instagram_story) and generates 4 thumbnails
+(YouTube, X share same prompt) via gpt-image-2 edit mode with reference
+image fallback. Results uploaded to content-studio-assets/{packetId}/.
+
+Default image model changed from gpt-image-1 to gpt-image-2."
+```
+
+---
+
+## Task 6: PostPickerModal 컴포넌트
+
+**Files:**
+
+- Create: `packages/web/app/admin/content-studio/PostPickerModal.tsx`
+
+- [ ] **Step 1: PostPickerModal 컴포넌트 작성**
+
+`packages/web/app/admin/content-studio/PostPickerModal.tsx`:
+
+```tsx
+"use client";
+
+import { useCallback, useEffect, useRef, useState } from "react";
+import { Loader2, Search, X } from "lucide-react";
+import { search } from "@/lib/api/generated/search/search";
+import { createBrowserClient } from "@/lib/supabase/client";
+
+type PostItem = {
+  id: string;
+  title: string;
+  thumbnail_url: string | null;
+  artist_name: string | null;
+  group_name: string | null;
+  view_count: number;
+  like_count: number;
+};
+
+type PostPickerModalProps = {
+  open: boolean;
+  onClose: () => void;
+  onSelect: (postId: string) => void;
+};
+
+export function PostPickerModal({
+  open,
+  onClose,
+  onSelect,
+}: PostPickerModalProps) {
+  const [query, setQuery] = useState("");
+  const [debouncedQuery, setDebouncedQuery] = useState("");
+  const [results, setResults] = useState<PostItem[]>([]);
+  const [loading, setLoading] = useState(false);
+  const [showManualInput, setShowManualInput] = useState(false);
+  const [manualId, setManualId] = useState("");
+  const inputRef = useRef<HTMLInputElement | null>(null);
+  const dialogRef = useRef<HTMLDialogElement | null>(null);
+
+  useEffect(() => {
+    const timer = setTimeout(() => setDebouncedQuery(query), 300);
+    return () => clearTimeout(timer);
+  }, [query]);
+
+  const fetchPopular = useCallback(async () => {
+    setLoading(true);
+    try {
+      const supabase = createBrowserClient();
+      const { data } = await supabase
+        .from("posts")
+        .select(
+          "id, title, thumbnail_url, artist_name, group_name, view_count, like_count",
+        )
+        .order("view_count", { ascending: false })
+        .limit(20);
+      setResults((data as PostItem[]) ?? []);
+    } catch {
+      setResults([]);
+    } finally {
+      setLoading(false);
+    }
+  }, []);
+
+  useEffect(() => {
+    if (!open) return;
+    if (debouncedQuery.trim().length === 0) {
+      void fetchPopular();
+      return;
+    }
+    setLoading(true);
+    search({ q: debouncedQuery, limit: 20 })
+      .then((res) => {
+        const items: PostItem[] = (res.data ?? []).map(
+          (hit: Record<string, unknown>) => ({
+            id: String(hit.id ?? ""),
+            title: String(hit.title ?? ""),
+            thumbnail_url: (hit.thumbnail_url as string) ?? null,
+            artist_name: (hit.artist_name as string) ?? null,
+            group_name: (hit.group_name as string) ?? null,
+            view_count: Number(hit.view_count ?? 0),
+            like_count: Number(hit.like_count ?? 0),
+          }),
+        );
+        setResults(items);
+      })
+      .catch(() => setResults([]))
+      .finally(() => setLoading(false));
+  }, [open, debouncedQuery, fetchPopular]);
+
+  useEffect(() => {
+    if (open) {
+      dialogRef.current?.showModal();
+      inputRef.current?.focus();
+    } else {
+      dialogRef.current?.close();
+    }
+  }, [open]);
+
+  if (!open) return null;
+
+  return (
+    <dialog
+      ref={dialogRef}
+      onClose={onClose}
+      className="fixed inset-0 z-50 m-auto h-[80vh] w-full max-w-xl rounded-lg border border-border bg-card p-0 shadow-xl backdrop:bg-black/50"
+    >
+      <div className="flex h-full flex-col">
+        <div className="flex items-center gap-2 border-b border-border px-4 py-3">
+          <Search className="h-4 w-4 text-muted-foreground" />
+          <input
+            ref={inputRef}
+            value={query}
+            onChange={(e) => setQuery(e.target.value)}
+            placeholder="Search posts..."
+            className="flex-1 bg-transparent text-sm text-foreground outline-none"
+          />
+          <button
+            type="button"
+            onClick={onClose}
+            className="text-muted-foreground hover:text-foreground"
+          >
+            <X className="h-4 w-4" />
+          </button>
+        </div>
+
+        <div className="flex-1 overflow-y-auto">
+          {loading ? (
+            <div className="flex items-center justify-center py-12">
+              <Loader2 className="h-5 w-5 animate-spin text-muted-foreground" />
+            </div>
+          ) : results.length === 0 ? (
+            <div className="py-12 text-center text-sm text-muted-foreground">
+              No posts found
+            </div>
+          ) : (
+            <div className="divide-y divide-border">
+              {results.map((post) => (
+                <button
+                  key={post.id}
+                  type="button"
+                  onClick={() => {
+                    onSelect(post.id);
+                    onClose();
+                  }}
+                  className="flex w-full items-center gap-3 px-4 py-3 text-left hover:bg-muted/60"
+                >
+                  {post.thumbnail_url ? (
+                    <img
+                      src={post.thumbnail_url}
+                      alt=""
+                      className="h-10 w-10 rounded object-cover"
+                    />
+                  ) : (
+                    <div className="h-10 w-10 rounded bg-muted" />
+                  )}
+                  <div className="min-w-0 flex-1">
+                    <div className="truncate text-sm font-medium text-foreground">
+                      {post.title}
+                    </div>
+                    <div className="text-xs text-muted-foreground">
+                      {[post.artist_name, post.group_name]
+                        .filter(Boolean)
+                        .join(" · ")}
+                    </div>
+                  </div>
+                  <div className="shrink-0 text-right text-xs text-muted-foreground">
+                    <div>{post.view_count.toLocaleString()} views</div>
+                    <div>{post.like_count.toLocaleString()} likes</div>
+                  </div>
+                </button>
+              ))}
+            </div>
+          )}
+        </div>
+
+        <div className="border-t border-border px-4 py-3">
+          {showManualInput ? (
+            <form
+              onSubmit={(e) => {
+                e.preventDefault();
+                if (manualId.trim()) {
+                  onSelect(manualId.trim());
+                  onClose();
+                }
+              }}
+              className="flex gap-2"
+            >
+              <input
+                value={manualId}
+                onChange={(e) => setManualId(e.target.value)}
+                placeholder="Enter post ID or UUID"
+                className="flex-1 rounded-md border border-input bg-background px-3 py-1.5 text-sm outline-none"
+              />
+              <button
+                type="submit"
+                disabled={!manualId.trim()}
+                className="rounded-md bg-foreground px-3 py-1.5 text-sm font-medium text-background disabled:opacity-50"
+              >
+                Select
+              </button>
+            </form>
+          ) : (
+            <button
+              type="button"
+              onClick={() => setShowManualInput(true)}
+              className="text-sm text-muted-foreground underline hover:text-foreground"
+            >
+              Enter ID manually
+            </button>
+          )}
+        </div>
+      </div>
+    </dialog>
+  );
+}
+```
+
+**주의**: `search()` 함수의 실제 파라미터 형식과 응답 타입은 `packages/web/lib/api/generated/search/search.ts`와 `SearchParams`/`SearchResponse` 타입을 확인하세요. 위 코드의 `{ q, limit }` 파라미터와 `res.data` 응답 매핑은 실제 API 스펙에 맞게 조정이 필요할 수 있습니다. `createBrowserClient`는 `@/lib/supabase/client`에서 import합니다 — 실제 경로를 확인하세요.
+
+- [ ] **Step 2: 타입 체크**
+
+Run: `cd packages/web && npx tsc --noEmit --pretty 2>&1 | grep "PostPickerModal"`
+Expected: 0 errors (import 전이므로 에러 없음)
+
+- [ ] **Step 3: 커밋**
+
+```bash
+git add packages/web/app/admin/content-studio/PostPickerModal.tsx
+git commit -m "feat(content-studio): add PostPickerModal component
+
+Meilisearch search with 300ms debounce + Supabase popular posts fallback
+(view_count descending, top 20). Shows thumbnail, title, artist/group,
+view/like counts. Manual ID input via fallback link at bottom."
+```
+
+---
+
+## Task 7: page.tsx 통합 — PostPickerModal + Research 제거 + 썸네일 UI
+
+**Files:**
+
+- Modify: `packages/web/app/admin/content-studio/page.tsx`
+
+- [ ] **Step 1: page.tsx에서 research 상태 및 ResearchPanel 제거**
+
+`packages/web/app/admin/content-studio/page.tsx`에서:
+
+1. import에서 `ResearchPanel`, `ResearchRun` 제거
+2. import에 `PostPickerModal` 추가
+3. state에서 `researchRun`, `researchWarning`, `useResearchInCopy` 삭제
+4. `handleCreatePacket`에서 research state 초기화 제거 (`setResearchRun(null)`, `setResearchWarning(null)`, `setUseResearchInCopy(false)`)
+5. `handleOpenPacket`에서 동일 research state 제거
+6. `handleGenerateVariants`에서 `researchContext`, `useResearchInCopy` API body 제거
+7. JSX에서 `<ResearchPanel ... />` 전체 제거
+8. `<AssetPanel>`에서 `researchRun={researchRun}`, `useResearchInCopy={useResearchInCopy}` props 제거
+9. `<ShortFormPanel>`에서 동일 props 제거
+
+- [ ] **Step 2: PostPickerModal 통합 + post ID 텍스트 입력 교체**
+
+```tsx
+// 추가 state:
+const [pickerOpen, setPickerOpen] = useState(false);
+const [keywords, setKeywords] = useState<string[]>([]);
+const [imagePrompts, setImagePrompts] = useState<{
+  youtube: string;
+  instagram_feed: string;
+  instagram_story: string;
+} | null>(null);
+const [thumbnails, setThumbnails] = useState<Record<string, string>>({});
+const [thumbnailLoading, setThumbnailLoading] = useState(false);
+
+// handleSelectPost: PostPickerModal에서 선택 시 자동 packet 생성
+async function handleSelectPost(selectedPostId: string) {
+  setPostId(selectedPostId);
+  setState("loading");
+  setError(null);
+  setGovernance(null);
+  setGenerationWarning(null);
+  setVariants([]);
+  setKeywords([]);
+  setImagePrompts(null);
+  setThumbnails({});
+
+  try {
+    const data = await postJson<{ packet: ContentPacket }>(
+      "/api/v1/content/packets",
+      { postId: selectedPostId },
+    );
+    setPacket(data.packet);
+    await loadRecentPackets();
+    setState("idle");
+  } catch (err) {
+    setError(err instanceof Error ? err.message : "Failed to create packet");
+    setState("error");
+  }
+}
+```
+
+기존 form을 교체:
+
+```tsx
+<section className="rounded-lg border border-border bg-card p-5">
+  <div className="grid gap-3 md:grid-cols-[1fr_auto]">
+    <div>
+      <span className="mb-2 block text-sm font-medium text-foreground">
+        Post
+      </span>
+      <button
+        type="button"
+        onClick={() => setPickerOpen(true)}
+        className="h-10 w-full rounded-md border border-input bg-background px-3 text-left text-sm text-foreground hover:border-foreground"
+      >
+        {postId ? postId : "Select a post..."}
+      </button>
+    </div>
+    <div className="flex items-end gap-2">
+      <select
+        value={generationMode}
+        onChange={(event) =>
+          setGenerationMode(event.target.value as ContentGenerationMode)
+        }
+        className="h-10 rounded-md border border-border bg-background px-3 text-sm text-foreground"
+        aria-label="Generation mode"
+      >
+        <option value="template">Template</option>
+        <option value="hybrid">Hybrid LLM</option>
+        <option value="llm">LLM</option>
+      </select>
+      <button
+        type="button"
+        disabled={!canGenerate}
+        onClick={handleGenerateVariants}
+        className="inline-flex h-10 items-center gap-2 rounded-md bg-foreground px-4 text-sm font-medium text-background disabled:cursor-not-allowed disabled:opacity-50"
+      >
+        {state === "loading" ? (
+          <Loader2 className="h-4 w-4 animate-spin" />
+        ) : (
+          <Sparkles className="h-4 w-4" />
+        )}
+        Generate
+      </button>
+    </div>
+  </div>
+
+  {error && (
+    <div className="mt-4 rounded-md border border-red-200 bg-red-50 px-3 py-2 text-sm text-red-700 dark:border-red-900 dark:bg-red-950/30 dark:text-red-300">
+      {error}
+    </div>
+  )}
+  {generationWarning && (
+    <div className="mt-4 rounded-md border border-amber-200 bg-amber-50 px-3 py-2 text-sm text-amber-700 dark:border-amber-900 dark:bg-amber-950/30 dark:text-amber-300">
+      {generationWarning}
+    </div>
+  )}
+</section>
+
+<PostPickerModal
+  open={pickerOpen}
+  onClose={() => setPickerOpen(false)}
+  onSelect={handleSelectPost}
+/>
+```
+
+- [ ] **Step 3: 썸네일 생성 UI 추가**
+
+variants 아래, GovernancePanel 위에 썸네일 섹션 추가:
+
+```tsx
+{
+  imagePrompts && (
+    <section className="rounded-lg border border-border bg-card p-5">
+      <h2 className="mb-4 text-base font-semibold text-foreground">
+        Thumbnail Generation
+      </h2>
+      <div className="grid gap-4 md:grid-cols-3">
+        {(["youtube", "instagram_feed", "instagram_story"] as const).map(
+          (channel) => (
+            <label key={channel} className="block">
+              <span className="mb-1 block text-xs font-medium uppercase text-muted-foreground">
+                {channel.replace("_", " ")}
+              </span>
+              <textarea
+                value={imagePrompts[channel]}
+                onChange={(e) =>
+                  setImagePrompts((prev) =>
+                    prev ? { ...prev, [channel]: e.target.value } : prev,
+                  )
+                }
+                rows={3}
+                className="w-full rounded-md border border-input bg-background px-3 py-2 text-sm outline-none"
+              />
+            </label>
+          ),
+        )}
+      </div>
+      <button
+        type="button"
+        disabled={thumbnailLoading || !packet}
+        onClick={async () => {
+          if (!packet || !imagePrompts) return;
+          setThumbnailLoading(true);
+          try {
+            const result = await postJson<{
+              thumbnails: Record<string, string>;
+              failures: Array<{ channel: string; error: string }>;
+            }>(`/api/v1/content/packets/${packet.id}/generate-thumbnails`, {
+              imagePrompts,
+              referenceImageUrl: packet.sourceImage || undefined,
+            });
+            setThumbnails(result.thumbnails);
+            if (result.failures.length > 0) {
+              setGenerationWarning(
+                `Thumbnail failures: ${result.failures.map((f) => `${f.channel}: ${f.error}`).join(", ")}`,
+              );
+            }
+          } catch (err) {
+            setError(
+              err instanceof Error
+                ? err.message
+                : "Thumbnail generation failed",
+            );
+          } finally {
+            setThumbnailLoading(false);
+          }
+        }}
+        className="mt-4 inline-flex items-center gap-2 rounded-md bg-foreground px-4 py-2 text-sm font-medium text-background disabled:opacity-50"
+      >
+        {thumbnailLoading ? (
+          <Loader2 className="h-4 w-4 animate-spin" />
+        ) : (
+          <Sparkles className="h-4 w-4" />
+        )}
+        Generate Thumbnails
+      </button>
+
+      {Object.keys(thumbnails).length > 0 && (
+        <div className="mt-4 grid grid-cols-2 gap-3">
+          {Object.entries(thumbnails).map(([channel, url]) => (
+            <div
+              key={channel}
+              className="overflow-hidden rounded-md border border-border"
+            >
+              <img src={url} alt={channel} className="w-full" />
+              <div className="px-2 py-1 text-xs font-medium uppercase text-muted-foreground">
+                {channel.replace("_", " ")}
+              </div>
+            </div>
+          ))}
+        </div>
+      )}
+    </section>
+  );
+}
+```
+
+- [ ] **Step 4: handleGenerateVariants에서 keywords/imagePrompts 상태 업데이트**
+
+`handleGenerateVariants` 함수를 수정하여 응답에서 `keywords`와 `imagePrompts`를 추출:
+
+```typescript
+async function handleGenerateVariants() {
+  if (!packet) return;
+  setState("loading");
+  setError(null);
+  setGenerationWarning(null);
+
+  try {
+    const variantData = await postJson<{
+      variants: ContentVariant[];
+      keywords?: string[];
+      imagePrompts?: {
+        youtube: string;
+        instagram_feed: string;
+        instagram_story: string;
+      };
+      mode?: ContentGenerationMode;
+      warning?: string | null;
+    }>(`/api/v1/content/packets/${packet.id}/generate-variants`, {
+      packet,
+      mode: generationMode,
+      locale: "ko-KR",
+    });
+    setVariants(variantData.variants);
+    setGenerationWarning(variantData.warning ?? null);
+    if (variantData.keywords) setKeywords(variantData.keywords);
+    if (variantData.imagePrompts) setImagePrompts(variantData.imagePrompts);
+
+    const reviewData = await postJson<{ result: GovernanceResult }>(
+      "/api/v1/content/variants/draft/review",
+      { packet, variants: variantData.variants },
+    );
+    setGovernance(reviewData.result);
+    setState("idle");
+    await loadRecentPackets();
+  } catch (err) {
+    setError(
+      err instanceof Error ? err.message : "Failed to generate variants",
+    );
+    setState("error");
+  }
+}
+```
+
+**주의**: generate-variants API route도 `keywords`와 `imagePrompts`를 응답에 포함하도록 수정해야 합니다. route handler에서 `generateUnifiedContent()` 호출로 교체하고 응답에 해당 필드를 추가하세요.
+
+- [ ] **Step 5: 키워드 표시 UI (variants 위에)**
+
+```tsx
+{
+  keywords.length > 0 && (
+    <section className="rounded-lg border border-border bg-card p-5">
+      <h2 className="mb-2 text-sm font-semibold text-foreground">Keywords</h2>
+      <div className="flex flex-wrap gap-1.5">
+        {keywords.map((kw) => (
+          <span
+            key={kw}
+            className="rounded-full bg-muted px-2.5 py-0.5 text-xs text-foreground"
+          >
+            {kw}
+          </span>
+        ))}
+      </div>
+    </section>
+  );
+}
+```
+
+- [ ] **Step 6: 타입 체크**
+
+Run: `cd packages/web && npx tsc --noEmit --pretty 2>&1 | head -30`
+Expected: 0 errors
+
+- [ ] **Step 7: 커밋**
+
+```bash
+git add packages/web/app/admin/content-studio/page.tsx
+git commit -m "feat(content-studio): integrate PostPickerModal, keywords, thumbnails
+
+Replace text input with PostPickerModal for post selection. Remove all
+research state (researchRun, researchWarning, useResearchInCopy) and
+ResearchPanel. Add keywords display, editable image prompts, and 4-channel
+thumbnail generation UI (2x2 grid preview). AssetPanel/ShortFormPanel
+research props removed."
+```
+
+---
+
+## Task 8: generate-variants route 업데이트
+
+**Files:**
+
+- Modify: `packages/web/app/api/v1/content/packets/[id]/generate-variants/route.ts`
+
+- [ ] **Step 1: route handler에서 generateUnifiedContent 사용**
+
+`packages/web/app/api/v1/content/packets/[id]/generate-variants/route.ts`에서:
+
+1. `generateVariantsWithMode` import을 `generateUnifiedContent`로 교체
+2. 응답에 `keywords`, `imagePrompts` 포함
+
+```typescript
+import { generateUnifiedContent } from "@/lib/content-studio/llm-generation";
+
+// POST handler 내부:
+const result = await generateUnifiedContent({
+  packet: parsed.data.packet,
+  mode: parsed.data.mode,
+  locale: parsed.data.locale,
+});
+
+return NextResponse.json({
+  variants: result.variants,
+  keywords: result.keywords,
+  imagePrompts: result.imagePrompts,
+  mode: result.mode,
+  warning: result.warning,
+});
+```
+
+- [ ] **Step 2: 타입 체크**
+
+Run: `cd packages/web && npx tsc --noEmit --pretty 2>&1 | head -20`
+Expected: 0 errors
+
+- [ ] **Step 3: 커밋**
+
+```bash
+git add packages/web/app/api/v1/content/packets/\[id\]/generate-variants/
+git commit -m "feat(content-studio): switch generate-variants to unified content
+
+Use generateUnifiedContent() instead of generateVariantsWithMode().
+Response now includes keywords and imagePrompts alongside variants."
+```
+
+---
+
+## Task 9: 테스트 정비 + 전체 검증
+
+**Files:**
+
+- Modify: `packages/web/lib/content-studio/__tests__/content-studio.test.ts`
+- Modify: `packages/web/lib/content-studio/__tests__/llm-generation.test.ts`
+
+- [ ] **Step 1: content-studio.test.ts에서 research 관련 테스트 정리**
+
+1. `externalEvidence` 필드를 사용하는 테스트 제거 또는 수정
+2. `researchProvenance`, `claims`, `missingFacts` 참조 제거
+3. governance 테스트에서 research-only 규칙 (`RESEARCH_COPY_WITHOUT_EVIDENCE`, `UNSUPPORTED_CLAIM`, `TREND_WITHOUT_SOURCE`, `LOW_CONFIDENCE_RESEARCH_CLAIM`) 관련 테스트 케이스 삭제
+4. mock data에서 `externalEvidence`, `researchProvenance`, `claims`, `missingFacts` 필드 제거
+
+- [ ] **Step 2: llm-generation.test.ts에서 research 테스트 정리**
+
+1. `researchContext`, `useResearchInCopy` 파라미터를 사용하는 테스트 제거
+2. `generateVariantsWithMode` 테스트에서 research 관련 assertion 제거
+3. `generateUnifiedContent` 테스트 추가 (Task 4에서 이미 작성한 것 확인)
+
+- [ ] **Step 3: 전체 테스트 실행**
+
+Run: `cd packages/web && npx vitest run lib/content-studio/__tests__/ 2>&1`
+Expected: 모든 테스트 PASS
+
+- [ ] **Step 4: 전체 타입 체크**
+
+Run: `cd packages/web && npx tsc --noEmit --pretty 2>&1`
+Expected: 0 errors
+
+- [ ] **Step 5: 빌드 테스트**
+
+Run: `cd packages/web && npx next build 2>&1 | tail -20`
+Expected: Build successful
+
+- [ ] **Step 6: 커밋**
+
+```bash
+git add packages/web/lib/content-studio/__tests__/
+git commit -m "test(content-studio): update tests for pipeline v2
+
+Remove research-related test cases (governance research rules,
+researchContext parameters, researchProvenance assertions).
+Add generateUnifiedContent and extractRuleKeywords tests."
+```
+
+---
+
+## Execution Notes
+
+### 의존성 순서
+
+Task 1 (UUID) → Task 2 (스키마) → Task 3 (research 삭제) → Task 4 (통합 LLM) → Task 5 (썸네일 API) → Task 6 (PostPickerModal, 독립) → Task 7 (page 통합) → Task 8 (route 업데이트) → Task 9 (테스트/검증)
+
+### 병렬 가능 태스크
+
+- Task 5 (썸네일 API)와 Task 6 (PostPickerModal)은 독립적이므로 병렬 실행 가능
+
+### 주의사항
+
+- `search()` 함수의 실제 `SearchParams` 타입을 확인하고 PostPickerModal의 호출을 조정하세요
+- `createBrowserClient`의 실제 import 경로를 확인하세요
+- `llm-client.ts`의 `generateContentVariantsWithOpenAI` 함수가 unified schema를 사용하도록 JSON schema 교체 필요
+- `uploadToStorage` 함수가 `openai-client.ts` 내 private이면 export 추가 필요
+- AssetPanel/ShortFormPanel의 API 호출 body에서도 `researchRun`, `useResearchInCopy` 제거 확인
+- 관련 API route들(`assets/plan`, `short-form/plan`)에서도 request schema가 변경되었으므로 해당 route handler 확인 필요