@autoaiai/aicut 0.1.11 → 0.1.13

package/skills/narrator-ai-cli/references/workflows.md

@@ -1,332 +0,0 @@
-# Workflows — Detailed Steps
-
-> ← Back to [SKILL.md](../SKILL.md)
-
-Full parameter tables and JSON request examples for both pipelines. The high-level flow lives in **SKILL.md § Fast Path / Standard Path**. Resource selection (material, BGM, dubbing, template) is covered in `resources.md`.
-
----
-
-## Fast Path — Detailed Steps
-
-### Step 0 — Find source material & determine `target_mode`
-
-Decision flow (see `resources.md` for the material-list command and programmatic search):
-
-1. **Found in pre-built materials** → construct `confirmed_movie_json` from material fields (mapping in `resources.md`). Ask user which `target_mode`:
-   - **`target_mode=1` 纯解说 (Pure narration)**: only metadata (title, synopsis, cast). Faster, no subtitle processing. Best when narration can be written from plot knowledge alone. **No `episodes_data`.**
-   - **`target_mode=2` 原声混剪 (Original mix)**: uses the actual subtitle track (`srt_file_id`) to align narration with original dialogue. More authentic. Requires `episodes_data` with `srt_oss_key = material.srt_file_id`.
-2. **Not found, known movie/drama** → run `task search-movie` (below) → `target_mode=1` with the returned `confirmed_movie_json`. **No `episodes_data`.**
-3. **Not found, user provides their own SRT (known movie)** → run `task search-movie` for `confirmed_movie_json` → `target_mode=2`. Use uploaded SRT as `srt_oss_key` in `episodes_data`.
-4. **Obscure / new drama, user provides SRT** → `target_mode=3`. `confirmed_movie_json` optional. Use uploaded SRT in `episodes_data`.
-
-**`search-movie` command** (only for cases 2 and 3 above; never fabricate its output):
-
-```bash
-narrator-ai-cli task search-movie "飞驰人生" --json
-```
-
-Returns up to 3 results, each:
-
-```json
-{
-  "title": "string",
-  "local_title": "string",
-  "year": "string",
-  "director": "string",
-  "stars": ["string"],
-  "genre": "string",
-  "summary": "string"
-}
-```
-
-⚠️ May take **60+ seconds** (Gradio backend). Results cached 24h.
-
-### Step 1 — fast-writing
-
-Construct the request body based on the `target_mode` chosen in Step 0:
-
-```bash
-narrator-ai-cli task create fast-writing --json -d @request.json
-```
-
-**Request shape by case** (use `-d @request.json` for readability; the four cases differ only in `target_mode`, `confirmed_movie_json`, and `episodes_data`):
-
-| Case | `target_mode` | `confirmed_movie_json` | `episodes_data` |
-|---|---|---|---|
-| A1 — material found, pure narration | `"1"` | mapped from material fields | omit |
-| A2 — material found, original mix | `"2"` | mapped from material fields | `[{srt_oss_key: <material.srt_file_id>, num: 1}]` |
-| B — known movie via search-movie, pure narration | `"1"` | from `search-movie` result | omit |
-| C — known movie + user SRT, original mix | `"2"` | from `search-movie` result | `[{srt_oss_key: <uploaded srt file_id>, num: 1}]` |
-| D — obscure/new drama, user SRT | `"3"` | optional (omit if unknown) | `[{srt_oss_key: <uploaded srt file_id>, num: 1}]` |
-
-**Example — Case D** (obscure drama):
-
-```json
-{
-  "learning_model_id": "<from narration-styles>",
-  "target_mode": "3",
-  "playlet_name": "<drama name>",
-  "episodes_data": [{"srt_oss_key": "<uploaded srt file_id>", "num": 1}],
-  "model": "flash"
-}
-```
-
-**Full parameter table:**
-
-| Parameter | Type | Required | Default | Description |
-|---|---|---|---|---|
-| `learning_model_id` | str | Exactly one (mutually exclusive with `learning_srt`) | - | Style model ID from a pre-built template or popular-learning result. **Do not provide both.** |
-| `learning_srt` | str | Exactly one (mutually exclusive with `learning_model_id`) | - | Reference SRT file_id. Use only when no template / popular-learning model is available. **Do not provide both.** |
-| `target_mode` | str | Yes | - | `"1"` Pure narration (热门影视), `"2"` Original mix (原声混剪), `"3"` New drama (冷门/新剧) |
-| `playlet_name` | str | Yes | - | Movie / drama name |
-| `playlet_num` | str | No | `"1"` | Episode/part number. Use `"1"` for single-episode; increment for multi-part |
-| `confirmed_movie_json` | obj | mode=1, 2; optional mode=3 | - | From material data (material found) or `search-movie` result. **Never fabricate.** |
-| `episodes_data` | list | mode=2, 3 | - | For fast-writing: `[{srt_oss_key, num}]`. For fast-clip-data: `[{video_oss_key, srt_oss_key, negative_oss_key, num}]` — video fields added at the clip-data step |
-| `model` | str | No | `"pro"` | `"pro"` (higher quality, 15 pts/char) or `"flash"` (faster, 5 pts/char) |
-| `language` | str | No | `"Chinese (中文)"` | Output language for the narration script. **Must match the selected dubbing voice language.** If voice is non-Chinese, set this explicitly — never leave at default |
-| `perspective` | str | No | `"third_person"` | `"first_person"` or `"third_person"` |
-| `target_character_name` | str | 1st person | - | Required when `perspective=first_person` |
-| `custom_script_result_path` | str | No | - | Custom script result path |
-| `webhook_url` | str | No | - | Async callback URL |
-| `webhook_token` | str | No | - | Callback authentication token |
-| `webhook_data` | str | No | - | Passthrough data for callback |
-
-**Output**: Creation response contains `data.task_id`. Poll `task query <task_id> --json` every 5 s until top-level `.status=2`. The completed response (flat object — see `operations.md` § Task Query Response Shape):
-
-```json
-{
-  "task_id": "<32-char hex>",
-  "task_order_num": "fast_writing_xxxxx",
-  "status": 2,
-  "completed_at": "2026-...",
-  "files": [{
-    "file_id": "<narration script file_id>",
-    "file_path": "user_data/.../narration.txt",
-    "suffix": ".txt"
-  }],
-  "results": {"tasks": [{"task_result": "user_data/.../narration.txt"}]}
-}
-```
-
-**Save**: `task_id` from the **creation response** (input to fast-clip-data), and `.files[0].file_id` from the **completed poll response** (also input to fast-clip-data as `file_id`).
-
-### Step 2 — fast-clip-data
-
-```bash
-narrator-ai-cli task create fast-clip-data --json -d '{
-  "task_id": "<task_id from step 1>",
-  "file_id": "<file_id from step 1>",
-  "bgm": "<bgm_id>",
-  "dubbing": "<voice_id>",
-  "dubbing_type": "<dubbing_type from selected voice>",
-  "episodes_data": [{
-    "video_oss_key": "<video_file_id>",
-    "srt_oss_key": "<srt_file_id>",
-    "negative_oss_key": "<video_file_id>",
-    "num": 1
-  }]
-}'
-```
-
-**Output**: Creation response → `data.task_id`. Poll until top-level `.status=2`. The completed response has **multiple `order_num`-shaped fields** — pick the right one (see `operations.md` § Task Query Response Shape for full structure):
-
-| Path in response | Format | Use as video-composing's `order_num`? |
-|---|---|---|
-| `.task_order_num` (top-level) | `fast_writing_clip_data_xxxxx` (prefixed string) | ✅ **YES — this is what video-composing wants** |
-| `.task_id` (top-level) | `2c95333519417a28b0d9d754fc6b8cc5` (32-char hex) | ❌ Submitting hex returns `10001 任务关联记录数据异常` |
-| `.results.order_info.order_num` (nested) | `script_xxxxx` | ❌ Billing-side identifier — also rejected by video-composing |
-
-### Step 3 — video-composing
-
-> ⚠️ **Field name collision warning**: video-composing's input parameter is named `order_num`, but its **value** must be the **top-level `.task_order_num`** field from Step 2's polled response — NOT `.task_id` (32-char hex) and NOT `.results.order_info.order_num` (`script_xxxxx`). Submitting either of the wrong values returns `10001`.
-
-```bash
-narrator-ai-cli task create video-composing --json -d '{
-  "order_num": "<value of top-level .task_order_num from Step 2 — looks like fast_writing_clip_data_xxxxx>"
-}'
-```
-
-**Required parameters**: `order_num` only. `order_num`'s value = fast-clip-data's top-level `.task_order_num` (prefixed string).
-
-**Output**: Creation returns `data.task_id`. Poll until top-level `.status=2`. Extract `video_url` from the nested results:
-
-```json
-{
-  "task_id": "<32-char hex>",
-  "task_order_num": "video_composing_xxxxx",
-  "status": 2,
-  "completed_at": "2026-...",
-  "results": {
-    "tasks": [{"video_url": "https://oss.example.com/.../output.mp4"}]
-  }
-}
-```
-
-`type_name` is `video_composing` (no BGM) or `video_composing_2` (with BGM); both return `video_url` at `.results.tasks[0].video_url`.
-
-### Step 4 (optional) — magic-video
-
-Only on explicit user request. See `magic-video.md`.
-
----
-
-## Standard Path — Detailed Steps
-
-### Step 0 — Find source material
-
-Same flow as Fast Path Step 0 but simpler — no `target_mode` decision. Either use a material's `video_file_id` + `srt_file_id`, or upload your own. Both feed directly into Step 2's `episodes_data`.
-
-### Step 1 — popular-learning (optional if using a pre-built template)
-
-```bash
-narrator-ai-cli task create popular-learning --json -d '{
-  "video_srt_path": "<srt_file_id from Step 0>",
-  "narrator_type": "movie",
-  "model_version": "advanced"
-}'
-```
-
-**`narrator_type`**: `短剧`, `电影`, `第一人称电影`, `多语种电影`, `第一人称多语种`, `movie`, `short_drama`, `first_person_movie`, `multilingual`, `first_person_multilingual`.
-
-**`model_version`**: `advanced` (高级版) or `standard` (标准版).
-
-**Output**: Poll until top-level `.status=2`. Parse `.results.tasks[0].task_result` (a JSON string) → `agent_unique_code` is the `learning_model_id`:
-
-```json
-{
-  "task_id": "<32-char hex>",
-  "status": 2,
-  "results": {
-    "tasks": [{
-      "task_result": "{\"agent_unique_code\": \"narrator-20251121160424-wjtOXO\"}"
-    }]
-  }
-}
-```
-
-→ `learning_model_id = "narrator-20251121160424-wjtOXO"`
-
-**Alternative (recommended)**: skip this step entirely — use a pre-built template `id` from `task narration-styles --json` directly as `learning_model_id`.
-
-### Step 2 — generate-writing
-
-`episodes_data` field sources:
-
-| `episodes_data` field | Source |
-|---|---|
-| `video_oss_key` | `video_file_id` from material (Step 0) or uploaded video `file_id` |
-| `negative_oss_key` | same as `video_oss_key` |
-| `srt_oss_key` | `srt_file_id` from material (Step 0) or uploaded SRT `file_id` |
-| `num` | episode number, starting from `1` |
-
-```bash
-narrator-ai-cli task create generate-writing --json -d '{
-  "learning_model_id": "<from step 1 or pre-built template>",
-  "playlet_name": "Movie Name",
-  "playlet_num": "1",
-  "target_platform": "douyin",
-  "vendor_requirements": "",
-  "target_character_name": "<protagonist name, or empty string>",
-  "episodes_data": [{
-    "video_oss_key": "<video_file_id>",
-    "srt_oss_key": "<srt_file_id>",
-    "negative_oss_key": "<video_file_id>",
-    "num": 1
-  }],
-  "refine_srt_gaps": false
-}'
-```
-
-**Required parameters:**
-
-| Parameter | Type | Required | Notes |
-|---|---|---|---|
-| `learning_model_id` | str | ✅ | From popular-learning result OR a pre-built template `id` |
-| `playlet_name` | str | ✅ | Movie / drama name |
-| `playlet_num` | str | ✅ | Episode number, e.g. `"1"` |
-| `target_platform` | str | ✅ | Distribution platform — `"douyin"`, `"youtube"`, etc. **API rejects request if missing.** Ask the user; default to `"douyin"` if unspecified |
-| `vendor_requirements` | str | ✅ | Custom vendor requirements. **Pass empty string `""` if none** — omitting the key returns `10001 vendor_requirements Field required` |
-| `target_character_name` | str | ✅ | Protagonist name. **Required even for third-person narration** — pass empty string `""` if not applicable. API will return a Pydantic `string_type` error if `null` / missing |
-| `episodes_data` | list | ✅ | Each entry **must contain all four** sub-fields: `video_oss_key`, `srt_oss_key`, `negative_oss_key`, `num`. Dropping `negative_oss_key` returns `第N个剧集缺少必要字段: negative_oss_key` |
-
-**Optional parameters:**
-
-| Parameter | Type | Default | Notes |
-|---|---|---|---|
-| `refine_srt_gaps` | bool | `false` | Enables AI scene analysis. **Only set to `true` on explicit user request** |
-| `language` | str | `"Chinese (中文)"` | Output language for the narration script — see Language linkage below |
-
-> ⚠️ **Common 10001 errors and their fix** (in order — fix one and the next surfaces):
-> 1. `target_platform Field required` → add `"target_platform": "douyin"` (or other platform)
-> 2. `vendor_requirements Field required` → add `"vendor_requirements": ""`
-> 3. `target_character_name ... Input should be a valid string` → add `"target_character_name": "<name or empty string>"`
-> 4. `第N个剧集缺少必要字段: negative_oss_key` → ensure every `episodes_data` entry contains all of `video_oss_key`, `srt_oss_key`, `negative_oss_key`, `num`
-
-> ⚠️ **Language linkage**: If the selected dubbing voice is non-Chinese, add `"language": "<target language>"` to this request. Default is Chinese; do NOT omit when using a non-Chinese voice. (See SKILL.md § Agent Rules — "Honor the language chain".)
-
-**Output**: Poll until top-level `.status=2`. The completed response (flat — see `operations.md` § Task Query Response Shape):
-
-```json
-{
-  "task_id": "<32-char hex>",
-  "task_order_num": "generate_writing_19c237b0_4ee66d",
-  "status": 2,
-  "completed_at": "2026-...",
-  "files": [{
-    "file_id": "<narration script file_id>",
-    "file_path": "user_data/.../narration.txt"
-  }],
-  "results": {
-    "tasks": [{"task_result": "user_data/.../narration.txt"}],
-    "order_info": {"order_num": "script_69269bfc_GfVEgA"}
-  }
-}
-```
-
-**Save (TWO different fields, both needed):**
-- `task_id` from the **creation response** — used to poll Step 2 to completion
-- **Top-level `.task_order_num`** from the **completed poll response** — required as `order_num` input for clip-data in Step 3 (e.g. `generate_writing_19c237b0_4ee66d`). ⚠️ Do **not** use `.results.order_info.order_num` (`script_xxxxx`); that's a separate billing-side identifier and is rejected by clip-data.
-
-### Step 3 — clip-data
-
-> ⚠️ **Input field name differs from fast-clip-data**: `clip-data` takes **`order_num`** (value = generate-writing's `task_order_num`, the prefixed string). `fast-clip-data` in Fast Path takes `task_id` instead. Same conceptual chaining, different parameter name. Submitting `task_id` to `clip-data` returns `10001 请输入合成时的任务ID或者订单号`.
-
-```bash
-narrator-ai-cli task create clip-data --json -d '{
-  "order_num": "<task_order_num from Step 2 — looks like generate_writing_xxxxx>",
-  "bgm": "<bgm_id>",
-  "dubbing": "<voice_id>",
-  "dubbing_type": "<dubbing_type from selected voice>"
-}'
-```
-
-**Output**: Creation returns `data.task_id`. Poll until top-level `.status=2`. The clip-data task record's `task_order_num` looks like `generate_clip_data_xxxxx`. **However, this is NOT what video-composing wants in Standard Path** — see the warning in Step 4 below.
-
-### Step 4 — video-composing
-
-> ⚠️ **Path asymmetry — Standard Path video-composing uses generate-writing's order_num, NOT clip-data's**:
-> - **Fast Path** (`fast-clip-data` → `video-composing`): pass `fast-clip-data`'s `task_order_num` (a combined `fast_writing_clip_data_xxxxx` that carries both contexts).
-> - **Standard Path** (`generate-writing` → `clip-data` → `video-composing`): pass **`generate-writing`'s `task_order_num`** (`generate_writing_xxxxx`). The `clip-data` step's own `task_order_num` (`generate_clip_data_xxxxx`) returns `10001 任务关联记录信息缺失: task_id=None, order_num=generate_clip_data_xxxxx`.
->
-> Mental model: in Standard Path, video-composing chains back to the **writing** step (clip-data is a side-effect that must complete first but is not what video-composing keys off). In Fast Path, fast-clip-data is a combined writing+clip operation, so its order is the right anchor.
-
-```bash
-narrator-ai-cli task create video-composing --json -d '{
-  "order_num": "<generate-writing task_order_num — looks like generate_writing_xxxxx>",
-  "bgm": "<bgm_id — same as Step 3>",
-  "dubbing": "<voice_id — same as Step 3>",
-  "dubbing_type": "<dubbing_type — same as Step 3>"
-}'
-```
-
-**Required parameters** (per `task types` definition): `order_num`, `bgm`, `dubbing`, `dubbing_type` — all four. Re-pass the same `bgm` / `dubbing` / `dubbing_type` values used in Step 3's clip-data. Submitting only `order_num` returns `10001 查询解说工程任务结果失败` even when both upstream tasks are at `status=2`.
-
-Pre-flight checklist:
-1. `generate-writing` (Step 2) has reached top-level `.status=2` and you've recorded its `task_order_num`.
-2. `clip-data` (Step 3) has reached top-level `.status=2` (must be done — but its `task_order_num` is **not** the value you pass).
-3. Pass `generate-writing`'s `task_order_num` (Step 2), not clip-data's (Step 3).
-4. Re-pass `bgm` / `dubbing` / `dubbing_type` — the API does not inherit them from clip-data.
-
-### Step 5 (optional) — magic-video
-
-Only on explicit user request. See `magic-video.md`.

package/skills/newcut-video-clipping/SKILL.md

@@ -1,194 +0,0 @@
----
-name: newcut-video-clipping
-description: "Transcribe long Chinese videos, semantically select complete standalone highlights, create real-quote hooks, clean pauses and reviewed speech mistakes, generate single-line subtitles and strong viewpoint titles, render clips with FFmpeg, and produce QC artifacts. Use when a user asks to cut livestreams, interviews, podcasts, talks, or other long Chinese videos into publishable short clips. Runs in AICUT: the film is film.html, a page per shot, drawn by `aicut look` and `aicut render`."
-license: MIT
-source: "hahadu4520/newcut-video-clipping-skill"
----
-
-# NewCut Video Clipping
-
-## Running it in AICUT
-
-AICUT's film is `film.html`: it cuts pages, footage, pictures and sound together on tracks. A page is a web page that
-sets `window.film = { frame(t) }` and draws that moment — no framework, no build step, no composition object, no
-`registerRoot`. Five shots are five pages cut together, not one page that changes scene, so each shot stays trimmable
-and replaceable in Studio.
-
-```sh
-npx -y @autoaiai/aicut@latest          # the manual: the format and the commands, always current
-npx -y @autoaiai/aicut@latest open .   # make this folder a film and print Studio's address
-npx -y @autoaiai/aicut@latest look     # a contact sheet of the whole film, its sound, and a determinism check
-npx -y @autoaiai/aicut@latest look 4.5 # one frame, full size — look at the PNG, not at the code
-npx -y @autoaiai/aicut@latest render   # the MP4, beside film.html
-```
-
-Run the manual before the first film: it is authoritative for the format, and this skill is about the method above it.
-
-Where the words below name another program's part, it becomes this:
-
-| elsewhere | in AICUT |
-|---|---|
-| composition, timeline, project file | `film.html`; tracks and clips inside it |
-| a scene, a sequence, a slide | a page (`pages/<name>.html`) cut onto a track |
-| `useCurrentFrame()`, `interpolate()` | `frame(t)` — the picture depends only on `t`, never on state or `Date.now()` |
-| a still, a screengrab | `aicut look <t>` |
-| `render` / `build` / `export` | `aicut render` (the person can also export from Studio) |
-| a dev server, a preview window | `aicut open` and the Studio address it prints |
-
-Hard rules that hold whatever the method says: the picture depends only on `t`; every value in the person's prompt is
-binding; look at the frames `look` writes before saying a film is done.
-
-
-Create polished clips through a reviewable pipeline. Prefer complete viewpoints over fixed durations and never select highlights by keyword scores.
-
-## Prerequisites
-
-Run the doctor before processing:
-
-```bash
-node <skill-root>/scripts/newcut.mjs doctor
-```
-```
-Required: Node.js 20+ and FFmpeg/FFprobe. For transcription, use an existing timestamped transcript, local Whisper, or user-provided Doubao/TOS credentials.
-
-For Doubao setup, read [provider-config.md](references/provider-config.md). Never print, copy, or commit credential values.
-```
-
-## Mandatory onboarding before requesting video
-
-For a first-time user, choose and verify the transcription setup before asking them to upload or provide a video. Do not wait until a video has already been uploaded. Do not run any media command during onboarding.
-
-Use this order exactly:
-
-```text
-choose transcription provider
--> complete provider setup
--> verify the setup
--> ask for the video
--> transcribe
-```
-First explain both choices in the user's language and ask them to choose:
-```
-```text
-开始前请选择转录方式：
-
-1. 本地 Whisper
-   - 不收取 API 费用，音视频不上传到第三方
-   - 首次使用需要下载模型文件
-   - 使用本机计算，长视频可能较慢
-   - 中文准确率和字级时间戳通常不如豆包稳定
-
-2. 豆包语音识别 2.0
-   - 中文和中英混合识别、字级时间戳更适合视频剪辑
-   - 需要注册火山引擎、开通服务并申请自己的 API Key
-   - 按火山引擎实际用量计费
-
-请选择“本地 Whisper”或“豆包语音”。
-```
-Wait for the user's answer. Do not silently select a provider. Skip this question only when the user supplied a transcript or explicitly chose a provider in the current request.
-
-After the choice:
-
-- For Whisper, explain that the first run downloads the selected model. Recommend `small` for a faster start or `medium` for better Chinese accuracy. Install or verify Whisper and FFmpeg first. Only after the environment is ready should you ask the user for a video.
-- For Doubao, open or link [provider-config.md](references/provider-config.md), guide the user through service activation and `.env.local`, and verify configuration before asking for a video. Never ask the user to paste a secret into chat.
-
-For Doubao onboarding, complete these checkpoints in order:
-
-1. The user has opened the Doubao Speech console and activated recording-file recognition 2.0.
-2. The user has created an API Key in the console.
-3. The user has filled `DOUBAO_API_KEY` and `DOUBAO_ASR_RESOURCE_ID` in a local `.env.local` file.
-4. Run `node <skill-root>/scripts/newcut.mjs doctor` without printing secret values.
-5. If required, run a credential connection check that does not need the user's production video.
-6. Confirm that configuration is ready.
-7. Only now ask the user to upload or provide the video file.
-
-If setup is incomplete, remain in onboarding. Do not ask for the production video merely to discover that credentials are missing.
-```
-
-## Workflow
-
-### 1. Transcribe and prepare semantic review
-
-With an existing transcript:
-
-```bash
-node <skill-root>/scripts/newcut.mjs process <video> \
-  --transcript <transcript.srt|json|txt> \
-  --output <job-dir>
-```
-```
-With Doubao ASR:
-```
-```bash
-node <skill-root>/scripts/newcut.mjs process <video> \
-  --asr-provider doubao \
-  --output <job-dir>
-```
-```
-The command creates the complete transcript, semantic review, source JSON, selection brief, and plan template. It does not select highlights by itself.
-```
-
-### 2. Read the complete transcript
-
-Read `<job-dir>/语义审阅稿.md` from beginning to end. Build a topic map and enumerate all complete viewpoints before selecting by count or preferred duration.
-
-Do not use:
-
-- Keyword scores.
-- Fixed sliding windows.
-- Silence as proof that a viewpoint ended.
-- Maximum duration during recall.
-- Rounded manual timestamps.
-
-### 3. Write the semantic plan
-
-Write `<job-dir>/semantic-plan.json` using raw utterance indexes. Every final clip must include:
-
-- A complete standalone argument.
-- Explicit `titleLines` with one or two semantically complete lines.
-- A hook copied from a real sentence inside the body.
-- Reviewed cleanup ranges for false starts, speech mistakes, or complex repetitions.
-- Timestamp-addressed subtitle corrections for confirmed ASR errors.
-
-Read [semantic-plan.md](references/semantic-plan.md) before authoring the plan.
-
-### 4. Render
-
-```bash
-node <skill-root>/scripts/newcut.mjs render \
-  --video <video> \
-  --transcript <job-dir>/semantic-source.json \
-  --plan <job-dir>/semantic-plan.json \
-  --output <clips-dir>
-```
-```
-The fixed render order is:
-```
-```text
-copied real-quote hook -> transition -> cleaned full body -> remapped subtitles
-```
-The hook remains in its original body position.
-```
-
-### 5. Quality check
-
-Verify each output is playable and inspect hook, middle, and ending frames. Confirm:
-
-- The viewpoint starts with enough context and ends after its final conclusion.
-- No first or last spoken character is cut.
-- Captions stay on one line and have no punctuation at either edge.
-- Caption chunks represent complete semantic phrases.
-- Title lines are semantically complete and stay inside the safe area.
-- Cleanup does not change the speaker's meaning or remove the original hook.
-- ASR corrections change subtitle display only, not spoken content.
-
-Read [quality-check.md](references/quality-check.md) for the full checklist.
-
-## Examples
-
-Read [examples.md](references/examples.md) for command and JSON examples. The repository intentionally ships no test media or transcript corpus.
-
-## Source
-
-Rewritten for AICUT from [`hahadu4520/newcut-video-clipping-skill`](https://github.com/hahadu4520/newcut-video-clipping-skill)
-(MIT), SKILL.md. The method is that skill's own; the way it runs is AICUT's.

package/skills/newcut-video-clipping/references/examples.md

@@ -1,59 +0,0 @@
-# Examples
-
-The skill repository intentionally contains no media, audio, subtitle, transcript, or private production fixture.
-
-## First-time onboarding
-
-Before requesting a video, the skill asks:
-
-```text
-开始前请选择转录方式：
-1. 本地 Whisper：免费、本地运行，首次需要下载模型。
-2. 豆包语音识别 2.0：中文和字级时间戳更稳定，需要自己的 API Key，并按量计费。
-
-请选择“本地 Whisper”或“豆包语音”。
-```
-
-If the user chooses Doubao, finish service activation, local `.env.local` configuration, and `doctor` verification first. Ask for the video only after configuration is ready.
-
-## Process a local video with an existing transcript
-
-```bash
-node <skill-root>/scripts/newcut.mjs process ./input.mp4 \
-  --transcript ./input.srt \
-  --output ./output/job-001
-```
-
-Expected artifact shape:
-
-```text
-output/job-001/
-├── source-info.json
-├── transcript.json
-├── semantic-source.json
-├── semantic-selection-brief.json
-├── semantic-plan.template.json
-├── 完整转录稿.md
-└── 语义审阅稿.md
-```
-
-After the agent writes `semantic-plan.json`:
-
-```bash
-node <skill-root>/scripts/newcut.mjs render \
-  --video ./input.mp4 \
-  --transcript ./output/job-001/semantic-source.json \
-  --plan ./output/job-001/semantic-plan.json \
-  --output ./output/job-001/clips
-```
-
-Expected render shape:
-
-```text
-output/job-001/clips/
-├── 01_viewpoint-title.mp4
-├── subtitles/
-│   └── 01_viewpoint-title.ass
-├── edit-plans.json
-└── resolved-clips.json
-```