Training Process Video

---
name: training-process-video
sheetId: "4.02"
description: "Post-process a livestream transcript into four parallel workstreams: an event recap Google Doc, a Q&A knowledge base diff, course lecture note proposals, and marketing content idea seeds."
category: training
project: training-process-video
platform: claude
status: Done
author: Peter Tu
input: "Raw livestream transcript text and a stream-config.md file identifying the training stream"
process: "4 parallel agents — event recap doc, Q&A knowledge base update, course outline enhancement, and content idea curation"
output: "Google Doc recap, Q&A diff entries, lecture note proposals, and content idea seeds"
synergy: []
type: agent
skills: ["training-event-recap", "training-qa-knowledge", "training-outline-enhancer", "training-idea-curator", "training-doc-admin"]
house-style: bound

---

# Process Livestream

Orchestrate four parallel agents to transform a livestream transcript into reusable course content, a knowledge base update, a published recap document, and new content ideas for the marketing pipeline.

---

## Step 0 — Stream selection

Scan the subdirectories of `<your-training-design-root>/` for folders that contain a `stream-config.md` file. Each such folder is an available training stream.

Display:

```
─────────────────────────────────────────
選擇訓練串流
─────────────────────────────────────────
請選擇此逐字稿所屬的訓練串流：

1️⃣ [folder name of first match]
2️⃣ [folder name of second match]
3️⃣ [folder name of third match]
（依掃描結果列出，數量不固定）
─────────────────────────────────────────
```

Wait for the user to type a number.

After selection:
- Store the absolute path to the selected stream directory as `STREAM_DIR` (e.g., `<your-training-design-root>/4.4 claude-code-for-operation`)
- Read `[STREAM_DIR]/stream-config.md` to load all stream configuration
- From the config, extract and store:
  - `course_outline`: filename of the course outline file (e.g., `4.4.1 course.md`)
  - `chapter_prefix`: prefix used for chapter files (e.g., `claude-code-ops`)
  - `chapter_count`: total number of chapter files
  - `recap_folder_id`: Google Drive folder ID for recap uploads
  - `roster_folder_id`: Google Drive folder ID for participant roster Sheets (`TODO` = skip Gate F)
  - `roster_input_folder_id`: Google Drive folder ID where a shortcut to the roster Sheet is created for GAS aggregation (`TODO` = skip shortcut creation)
  - `kit_tag_prefix`: Kit tag prefix for this stream, e.g. `[4] Claude Code` (`TODO` = skip Gate F.5)
  - All Google Doc/Sheet IDs listed in the Google Workspace Sync Map section of the config

Display: `✅ 已載入串流設定：[selected stream folder name]`

---

## Step 1 — Collect inputs

Ask the user for:
1. **Transcript file path**: path to the transcript `.md` file (can be anywhere on disk)
2. **Stream title**: short name for this session (e.g., "Claude Code 營運直播 no.1")
3. **Stream date**: YYYY-MM-DD format (default: today's date if not provided)

---

## Step 1a — Transcript intake (always runs)

The transcript file was already loaded into context by the user (via `@file` reference). This step only handles renaming and path storage — no file reading required.

**Compute standardized filename:**
1. Take `stream_title`, convert to lowercase
2. Replace spaces and special characters with hyphens; remove consecutive hyphens
3. Append `-[stream_date].cleaned.md`
4. Format: `[slug]-[YYYY-MM-DD].cleaned.md`

**Determine what action is needed:**

- If the source file is already in `[STREAM_DIR]/video-recording/` **and** its basename matches the standardized filename → no action needed; store the path as-is.
- If the source file is in `[STREAM_DIR]/video-recording/` but has a different name → rename in place:
  ```bash
  mv "[source_file_path]" "[STREAM_DIR]/video-recording/[standardized_filename]"
  ```
- If the source file is outside `[STREAM_DIR]/video-recording/` → copy it in (do not delete the original):
  ```bash
  cp "[source_file_path]" "[STREAM_DIR]/video-recording/[standardized_filename]"
  ```

Store the resulting absolute path as `FILE_PATH` and the filename as `FILE_NAME` for use in Step 10.

If the rename or copy fails, display the error and stop.

Display:
```
✅ 逐字稿就緒
   原始檔案：[source_file_path]
   標準化名稱：[FILE_NAME]
```

---

## Step 2 — Pre-flight check

Before launching agents, display:

```
─────────────────────────────────────────
直播後製 — 準備啟動
─────────────────────────────────────────
Stream : [title]
Date   : [date]
Length : [N] words / approx. [M] minutes of reading
─────────────────────────────────────────
將啟動 4 個平行 Agent：
  [1] Event Recap       → 回顧文章草稿（需審核後上傳 Google Drive）
  [2] Q&A KB            → Diff 報告（需審核後寫入）
  [3] Course Enhancer   → 講義提案（需審核後寫入）
  [4] Content Curator   → 內容點子提案（需審核後寫入 _matrix.md）
─────────────────────────────────────────
所有寫入與上傳均在審核後執行，不會自動發布。
繼續？ [Y/n]
```

Wait for confirmation. If the user types `n` or `no`, stop.

---

## Google Workspace Sync Map

After human approval at each gate, approved local file changes are also pushed to their corresponding Google Workspace documents. Use `<your-google-workspace-account>` as `user_google_email` for all MCP calls.

Read the Google Workspace Sync Map from `[STREAM_DIR]/stream-config.md`. Use the Doc/Sheet IDs listed there for all sync operations.

If any ID is `TODO`, skip that sync and note it in the summary.

**Sync procedure** (applies to Gates A, B, C, D):
1. Write approved content to the local file first.
2. Read the full updated local file.
3. Use `mcp__google-workspace__get_doc_content` to get the current Google Doc content.
4. Use `mcp__google-workspace__batch_update_doc` to replace the Google Doc body with the full updated local file content (delete existing text from index 1 to end, then insert at index 1).
5. For the Content Matrix Google Sheet (Gate D), use `mcp__google-workspace__modify_sheet_values` to append new rows to the appropriate sheet tab.
6. If any MCP sync call fails, report the error but do not block the workflow — the local file is the source of truth.

---

## Step 3 — Read reference files

Before launching agents, read the following files so their contents are available to pass to agents:

- `[FILE_PATH]` (the transcript — use this as `transcript` input for all agents)
- `[STREAM_DIR]/knowledge-base/qa-master.md` (KB contents for Agent 2)
- `[STREAM_DIR]/[course_outline from config]` (course outline for Agent 3)
- `[STREAM_DIR]/[chapter_prefix from config]-ch1.md` through `[STREAM_DIR]/[chapter_prefix from config]-ch[chapter_count].md` (chapter files for Agent 3)
- `[STREAM_DIR]/content-matrix/_matrix.md` (content matrix for Agent 4)

If any chapter file is missing, note it and pass an empty string for that chapter.
If `_matrix.md` cannot be read, note it and Agent 4 will treat all ideas as NEW.

If `[STREAM_DIR]/knowledge-base/qa-master.md` does not exist, prompt:

```
─────────────────────────────────────────
⚠️ 找不到 knowledge-base/qa-master.md
─────────────────────────────────────────
[B] Bootstrap — 依章節分類架構建立空白 qa-master.md（從 stream-config.md 讀取章節 taxonomy）
[S] 略過 Agent 2（本次執行不執行 Q&A 知識庫更新）
─────────────────────────────────────────
```

Wait for user input.
- **[B]**: Create `[STREAM_DIR]/knowledge-base/qa-master.md` with chapter taxonomy headers as defined in `stream-config.md`. Then proceed with Agent 2 normally.
- **[S]**: Skip Agent 2 entirely for this run. Set its status to `⏭ 已略過` on the status board.

---

## Step 4 — Launch 4 agents in parallel

Launch all four agents simultaneously using the Task tool. Show the status board immediately after launching:

```
─────────────────────────────────────────
Agent 狀態板
─────────────────────────────────────────
[1] Event Recap     ⏳ 執行中...
[2] Q&A KB          ⏳ 執行中...
[3] Course Enhancer ⏳ 執行中...
[4] Content Curator ⏳ 執行中...
─────────────────────────────────────────
```

**Agent 1 — Event Recap**
Instruction to agent (pass as prompt):
- Read `agents/training-event-recap.md` for your full instructions
- Inputs: transcript text, title, date
- Do NOT call any MCP tools — return formatted content only
- Return the structured result block described in your instructions

**Agent 2 — Q&A Knowledge Base**
Instruction to agent (pass as prompt):
- Read `agents/training-qa-knowledge.md` for your full instructions
- Inputs: transcript text, full contents of `[STREAM_DIR]/knowledge-base/qa-master.md`
- Local file reads only — no MCP writes
- Return the diff report described in your instructions

**Agent 3 — Course Outline Enhancer**
Instruction to agent (pass as prompt):
- Read `agents/training-outline-enhancer.md` for your full instructions
- Inputs: transcript text, course outline (`[STREAM_DIR]/[course_outline from config]`), all chapter files (`[STREAM_DIR]/[chapter_prefix]-ch1.md` through `[STREAM_DIR]/[chapter_prefix]-ch[chapter_count].md`)
- Local file reads only — no MCP writes
- Return the match report described in your instructions

**Agent 4 — Content Idea Curator**
Instruction to agent (pass as prompt):
- Read `agents/training-idea-curator.md` for your full instructions
- Inputs: transcript text, stream title, stream date, full contents of `[STREAM_DIR]/content-matrix/_matrix.md`
- Local file reads only — no MCP writes
- Return the idea report described in your instructions

---

## Step 5 — Collect results and update status board

As each agent completes, update the status board:

```
─────────────────────────────────────────
Agent 狀態板
─────────────────────────────────────────
[1] Event Recap     ✅ 完成 — 草稿已備妥，待審核上傳
[2] Q&A KB          ✅ 完成 — [N] 筆新增，[M] 筆重複
[3] Course Enhancer ✅ 完成 — [K] 段符合，[J] 段未匹配
[4] Content Curator ✅ 完成 — [L] 個點子符合，[P] 個略過
─────────────────────────────────────────
```

If any agent returned an error, show it clearly and continue with the remaining agents.

After all four complete (or fail), display a 1-paragraph plain-text summary for each.

---

## Step 6 — Gate A: Q&A Knowledge Base write

Display the MECE health check from Agent 2 first:

```
─────────────────────────────────────────
Gate A — MECE 健檢
─────────────────────────────────────────
[Display the full MECE HEALTH CHECK block from Agent 2]
─────────────────────────────────────────
```

If any MECE flags exist (CATEGORY_MISMATCH, OVERLAP_RISK, CATEGORY_BLOAT, ORPHAN_CATEGORY), prompt:

```
⚠️ 發現 [N] 個 MECE 問題，建議在寫入前處理。

[F] 逐筆修正（逐一顯示每個問題，讓你決定接受建議或略過）
[I] 忽略，直接進入寫入審核
```

Wait for user input. If [F]:
- For each flag, show the issue and the agent's suggestion. User types `y` to accept the suggested fix (update category / skip entry / merge), `n` to keep as-is.

Then display the full diff report and show:

```
─────────────────────────────────────────
Gate A — Q&A 知識庫寫入審核
─────────────────────────────────────────
新增：[N] 筆  |  Variant：[M] 筆  |  衝突：[K] 筆
（重複項目已自動略過）

[A] 全部寫入新增項目
[S] 逐筆選擇
[N] 略過，不寫入
─────────────────────────────────────────
```

Wait for user input.

- **[A]**: Insert all NEW entries into `[STREAM_DIR]/knowledge-base/qa-master.md` at their specified `insert_after` positions (the KB is ordered by chapter → topic; do NOT append to the end). If a new topic group is needed, add a `<!-- Topic: [name] -->` comment before the first entry in the group. For VARIANT and CONTRADICTS entries, ask separately: "是否更新現有 KB 條目？[Y/n]"
- **[S]**: Show each NEW entry one at a time (including its `insert_after` position). User types `y` to include, `n` to skip. After cycling through NEW entries, handle VARIANT/CONTRADICTS the same way.
- **[N]**: Skip all writes for Gate A.

After writing (if any), show: `已寫入 [N] 筆到 knowledge-base/qa-master.md`

**Google Sync:** If any entries were written to `[STREAM_DIR]/knowledge-base/qa-master.md`, sync the full updated file to the QA Master KB Google Doc using the ID from `stream-config.md` and the sync procedure described in the Google Workspace Sync Map section. Show: `☁️ 已同步至 Google Doc（QA Master KB）`

---

## Step 7 — Gate B: Lecture notes write

Display the match report from Agent 3, grouped by target chapter file.

Then show:

```
─────────────────────────────────────────
Gate B — 講義提案審核
─────────────────────────────────────────
符合段落：[N] 段，分布於 [K] 個章節檔案

[A] 全部寫入
[S] 逐段選擇
[N] 略過，不寫入
─────────────────────────────────────────
```

Wait for user input.

- **[A]**: Append all qualifying lecture note blocks to their respective chapter files.
- **[S]**: Show each proposal one at a time. User types `y` to append, `n` to skip.
- **[N]**: Skip all writes for Gate B.

For each approved block, append to the end of the correct chapter file (`[STREAM_DIR]/[chapter_prefix]-ch[N].md`).

After writing (if any), show: `已寫入 [N] 段到 [K] 個章節檔案`

**Google Sync:** For each chapter file that was modified, sync the full updated file to its corresponding Google Doc using the sync procedure and IDs from `stream-config.md`. Show: `☁️ 已同步 [K] 個章節至 Google Docs`

---

## Step 8 — Gate C: Unmatched content

If Agent 3 returned any unmatched segments, display them now:

```
─────────────────────────────────────────
Gate C — 未匹配內容處理
─────────────────────────────────────────
以下 [N] 段內容未能匹配現有課程章節：
```

For each unmatched segment, show it and prompt:

```
[段落描述]
Agent 建議：[add to outline / skip]

[1] 加入課程大綱（會標記在 [course_outline from config] 的建議清單）
[2] 加入章節檔末尾（選擇章節）
[3] 略過
```

Wait for user to type `1`, `2`, or `3`.

- **[1]**: Append a note to `[STREAM_DIR]/[course_outline from config]` at the end under a `## 直播內容建議` section (create if not exists).
- **[2]**: Ask which chapter, then append as a brief note at the end of that chapter file (`[STREAM_DIR]/[chapter_prefix]-ch[N].md`).
- **[3]**: Do nothing.

**Google Sync:** After all unmatched segments are processed, sync any modified files to their corresponding Google Docs using the sync procedure and IDs from `stream-config.md`. This may include `[course_outline from config]` and/or chapter files. Show: `☁️ 已同步未匹配內容至 Google Docs`

---

## Step 9 — Gate D: Content idea matrix write

If Agent 4 returned no qualifying ideas, skip this gate.

Otherwise, display the full idea report from Agent 4.

Then show:

```
─────────────────────────────────────────
Gate D — 內容點子寫入審核
─────────────────────────────────────────
符合提案：[N] 個  |  低分略過：[M] 個  |  重複略過：[K] 個
目標檔案：[STREAM_DIR]/content-matrix/_matrix.md

重複略過明細：
  - [pain summary] → 與 [existing-id] (global_index [NNN]) 重複
  - ...

低分略過明細：
  - [pain summary] | [X]/8 | 未通過：[failed checks]
  - ...

[A] 全部寫入符合提案
[S] 逐筆選擇
[N] 略過，不寫入
─────────────────────────────────────────
```

Wait for user input.

- **[A]**: Append all qualifying seed rows to `[STREAM_DIR]/content-matrix/_matrix.md` under the correct category table. Also append a new session entry to `[STREAM_DIR]/content-matrix/_sourcing-log.md` documenting this run: date, stream title, method "livestream transcript", count of ideas added.
- **[S]**: Show each proposal one at a time. User types `y` to include, `n` to skip. After all selections, write approved rows and log the session.
- **[N]**: Skip all writes for Gate D.

When appending to `[STREAM_DIR]/content-matrix/_matrix.md`:
- Add each approved row to the correct category table (prompt / skills / subagents / mcp / cot)
- Assign the `global_index` sequentially, incrementing from the current highest value in the matrix
- Update the Status Summary table at the bottom: increment `seed` count by the number of rows added, update `Last updated` date

When appending to `[STREAM_DIR]/content-matrix/_sourcing-log.md`:
- Add a new session block under `## Sessions`, newest first:

```markdown
### [YYYY-MM-DD] — [Stream Title]

**Method:** Livestream transcript
**Duration:** —
**Ideas added to matrix:** [N] ([list of IDs added])
**Validation runs:** 0 (all remain `seed`)
**Notes:** Extracted by training-idea-curator agent via /training-process-video.
```

After writing (if any), show: `已寫入 [N] 筆到 _matrix.md，並更新 _sourcing-log.md`

**Google Sync:** If any ideas were written to `_matrix.md`, also sync to the Content Matrix Google Sheet and sourcing-log sheet using the IDs from `stream-config.md`.

**Matrix sheet** (use the Content Matrix Sheet ID and sheet name `matrix` from `stream-config.md`):
- If the sheet has no header row, first write the header row to `matrix!A1:K1`:

| A | B | C | D | E | F | G | H | I | J | K |
|---|---|---|---|---|---|---|---|---|---|---|
| global_index | id | category | pain_point_raw | pain_resonance_score | solution_type | result_claim | status | file_path | sourced_from | notes |

- Append each approved seed row to the next empty row using `mcp__google-workspace__modify_sheet_values` with range `matrix!A[next_row]:K[next_row]`.

**Sourcing-log sheet** (use the same Content Matrix Sheet ID with sheet name `sourcing-log` from `stream-config.md`):
- If the sheet has no header row, first write the header row to `sourcing-log!A1:F1`:

| A | B | C | D | E | F |
|---|---|---|---|---|---|
| date | stream_title | method | ideas_added | idea_ids | notes |

- Append a session row with: date, stream title, "livestream transcript", count of ideas, comma-separated list of IDs, "Extracted by training-idea-curator agent via /training-process-video."

Show: `☁️ 已同步 [N] 筆至 Google Sheet（Content Matrix + Sourcing Log）`

---

## Step 9.5 — Gate E: Event Recap upload to Google Drive

If Agent 1 returned an error, skip this gate and note it in Step 10.

Otherwise, display the full formatted recap content from Agent 1 for review.

Then show:

```
─────────────────────────────────────────
Gate E — 活動回顧上傳審核
─────────────────────────────────────────
標題：[DOC_TITLE]
（以上為即將建立的 Google Doc 內容）

[Y] 確認上傳 Google Drive
[E] 編輯後再上傳（請直接修改上方內容並回覆 Y）
[N] 略過，不上傳
─────────────────────────────────────────
```

Wait for user input.

- **[Y]**: First, save a local copy of the approved recap content to `[STREAM_DIR]/[YYYY-MM-DD]-course-recap.md`. Then, read `recap_folder_id` from `stream-config.md`. If it is `TODO`, skip the Google Drive upload step and note it in the summary. Otherwise, create the Google Doc and move it to the recap folder using MCP:
  1. Write the approved recap content to `[STREAM_DIR]/[YYYY-MM-DD]-course-recap.md`
  2. `mcp__google-workspace__create_doc` — title: `[DOC_TITLE]`
  3. `mcp__google-workspace__batch_update_doc` — insert the approved content into the document body
  4. If `recap_folder_id` is not `TODO`: `mcp__google-workspace__update_drive_file` — move the doc to the recap folder by setting `add_parents: [recap_folder_id]` and `remove_parents: root`
  5. `mcp__google-workspace__get_drive_shareable_link` — retrieve shareable URL
  Store the URL as `RECAP_DOC_URL` and the local path as `RECAP_LOCAL_PATH` for use in Step 10.
  Show: `✅ 本地檔案：[RECAP_LOCAL_PATH]` and `✅ Google Doc 已建立並移至回顧資料夾：[RECAP_DOC_URL]`
- **[E]**: Wait for the user to provide edited content, then proceed as [Y] with the revised text.
- **[N]**: Skip upload and local save. Set `RECAP_DOC_URL` to "（略過）" and `RECAP_LOCAL_PATH` to "（略過）".

---

## Step 9.6 — Gate F: Participant roster upload

> This step handles the `.xlsx` participant roster downloaded from the event platform (e.g. Accupass/活動通). Accupass has no API — the user downloads the file manually and provides the path.

Read `roster_folder_id` from `stream-config.md`. If it is `TODO`, skip this gate, set `ROSTER_STATUS` to "略過（未設定 roster_folder_id）" and proceed to Step 9.7.

If no participant list path was provided in Step 1, ask:
```
─────────────────────────────────────────
Gate F — 參加名單歸檔
─────────────────────────────────────────
請提供活動平台下載的 .xlsx 檔案路徑：
（輸入路徑，或按 Enter 略過）
─────────────────────────────────────────
```

If user skips (empty Enter), set `ROSTER_STATUS` to "略過" and proceed to Step 9.7.

Otherwise, show:
```
─────────────────────────────────────────
Gate F — 參加名單歸檔
─────────────────────────────────────────
來源檔案：[xlsx 路徑]
上傳目標：[活動名稱]_[YYYY.M.DD]（Google Sheet）
目標資料夾：[roster_folder_id 對應資料夾]
[若 roster_input_folder_id 已設定] 同時建立捷徑至 input 資料夾（供 GAS 更新總表用）

[Y] 確認上傳
[N] 略過
─────────────────────────────────────────
```

**[Y] — execute (use Python + Drive API; MCP has no xlsx→Sheet conversion tool):**

1. Upload the xlsx and convert it to a Google Sheet in the `roster_folder_id` folder:
   ```python
   from googleapiclient.http import MediaFileUpload
   media = MediaFileUpload(xlsx_path,
       mimetype='application/vnd.openxmlformats-officedocument.spreadsheetml.sheet',
       resumable=True)
   file_meta = {
       'name': '[活動名稱]_[YYYY.M.DD]',
       'mimeType': 'application/vnd.google-apps.spreadsheet',
       'parents': ['[roster_folder_id]']
   }
   sheet = drive_service.files().create(body=file_meta, media_body=media, fields='id,webViewLink').execute()
   SHEET_ID = sheet['id']
   ```

2. If `roster_input_folder_id` in `stream-config.md` is not `TODO`, create a shortcut in that input folder (one per event):
   ```python
   shortcut_meta = {
       'name': '[活動名稱]_[YYYY.M.DD]',
       'mimeType': 'application/vnd.google-apps.shortcut',
       'shortcutDetails': {'targetId': SHEET_ID},
       'parents': ['[roster_input_folder_id]']
   }
   drive_service.files().create(body=shortcut_meta, fields='id').execute()
   ```

3. Set `ROSTER_STATUS` to "✅ 已上傳並建立捷徑"

Show:
```
✅ 參加名單已上傳為 Google Sheet：[webViewLink]
✅ 捷徑已建立至 input 資料夾（供 GAS 使用）
```

**[N]**: Set `ROSTER_STATUS` to "略過"

---

## Step 9.7 — Gate F.5: Upload subscribers to Kit

If `ROSTER_STATUS` is "略過" (Gate F was skipped), skip this gate and set `KIT_STATUS` to "略過".

Read `kit_tag_prefix` from `stream-config.md` (e.g. `[4] Claude Code`). If it is `TODO`, skip this gate and set `KIT_STATUS` to "略過（未設定 kit_tag_prefix）".

Otherwise, display:

```
─────────────────────────────────────────
Gate F.5 — 上傳訂閱者至 Kit
─────────────────────────────────────────
來源：參加名單 Google Sheet（[SHEET_ID]）
動作：新增至 Kit 訂閱者列表 + 打上活動標籤

[Y] 確認執行
[N] 略過
─────────────────────────────────────────
```

**[N]**: Set `KIT_STATUS` to "略過", proceed to Step 10.

**[Y] — execute:**

**1. Read the roster**
Read `SHEET_ID` with `mcp__google-workspace__get_drive_file_content` to get the CSV content.

Column mapping (Accupass format):
- Column index 2: 訂購人姓名 → Kit `first_name`
- Column index 3: 訂購人Email → Kit `email_address`

**2. Clean the data**
- Dedupe: keep the first row per email
- Filter: skip emails containing `privaterelay.appleid.com` (Apple private relay — undeliverable)

Display the cleaning result: 原始 N 筆 → 去重 X 筆 → 過濾 Y 筆 → 有效 Z 筆

**3. Resolve the Kit tag**
Use `mcp__kit__list_tags` to find the highest number among tags sharing the `kit_tag_prefix`, then +1 for the new tag number. Create it with `mcp__kit__create_tag` (returns the existing ID if the tag already exists).

**4. Bulk-create subscribers**
Call `mcp__kit__bulk_create_subscribers` in batches of ≤100 (upsert — existing subscribers are updated automatically).
Collect every valid `subscriber_id` — including those returned inside "Cannot override state" failure items. Those subscribers already exist and must still be tagged, otherwise they are silently missed.

**5. Bulk-tag subscribers**
Call `mcp__kit__bulk_tag_subscribers` in batches of ≤100 with the tag from step 3.

Set `KIT_STATUS` to "✅ [N] 筆已上傳，標籤：[tag_name]"

Show:
```
✅ Kit 訂閱者已更新：[N] 筆成功 | [M] 筆失敗
🏷 標籤：[tag_name]（ID: [tag_id]）
```

---

## Step 10 — Final summary and file archival

Display:

```
─────────────────────────────────────────
✅ Process Livestream 完成
─────────────────────────────────────────
Stream : [title] ([date])

📄 活動回顧
   本地檔案：[RECAP_LOCAL_PATH]
   Google Doc：[RECAP_DOC_URL]

🗂 Q&A 知識庫
   寫入：[N] 筆新增  |  更新：[M] 筆 Variant  |  略過：[K] 筆
   ☁️ Google Sync：[成功/略過/失敗]

📚 課程講義
   寫入：[N] 段  →  [list of chapter files modified]
   未匹配處理：[N] 段加入大綱建議  |  [M] 段略過
   ☁️ Google Sync：[K] 個文件已同步 / [失敗清單]

💡 內容點子
   寫入：[N] 個 seed → _matrix.md  |  略過：[M] 個
   ☁️ Google Sync：[成功/略過/失敗]

📋 參加名單
   [ROSTER_STATUS]

📧 Kit 訂閱者
   [KIT_STATUS]

─────────────────────────────────────────
```

After displaying the summary, call `agents/training-doc-admin.md` with:
- `action: move`
- `file_path`: the `FILE_PATH` stored from Step 1a
- `processed_transcript_dir`: `[STREAM_DIR]/processed-transcript/`

Display the result:
- On success: `📁 逐字稿已移至 processed-transcript/[FILE_NAME]`
- On error: surface the error message from the agent

---

## Error Handling

- If the admin agent returns `blocked` or `error` during transcript intake, stop the entire workflow — there is no transcript to process.
- If Agent 1 (Event Recap) returns an error, skip Gate E and note it in the final summary.
- If Agent 2 returns no Q&A pairs, skip Gate A entirely.
- If Agent 3 returns no qualifying segments, skip Gate B. Still run Gate C if unmatched segments exist.
- If a chapter file write fails, report the specific file and continue with others.
- If Agent 4 returns no qualifying ideas, skip Gate D entirely.
- If `_matrix.md` cannot be written, report the error — do not attempt to write `_sourcing-log.md` either.
- If any MCP call in Gate E fails, show the error and offer to retry or skip.
- Never silently swallow errors — always surface them in the status board.

## House style

Writing style is **not owned by this file**. The house voice lives in two Google Docs under
`[@] 寫作指南` (`12DBdFz3SK22ie9im_ThFMI7IBRXsTZsV`), read at runtime:

- 《[2.0] Zynkr 通用風格指南 House Voice》 `10bOIQwRm9Pxwgct4hlwCwK_B4Pipai1HqBPZKzyRHSE` —
  the universal core, plus the addendum for this surface
- 《[3.2] 禁用詞清單 Forbidden Words》 `1N5sHLP4qzmmhpCGsi6KElxi1z0MFe4QZ0Q_35T10Uyg`

Read both before producing client- or reader-facing text, and scan the draft against 《[3.2]》
before handing it over. If Drive is unreachable, say so in the output rather than proceeding
unchecked. Never re-implement either list inside this file.