@autoaiai/aicut 0.1.11 → 0.1.13

package/skills/openchatcut/references/getting-started.md

@@ -1,106 +0,0 @@
-# Connect OpenChatCut
-
-Use this flow when the user says “set up OpenChatCut”, when the `openchatcut`
-MCP server is missing, or when its endpoint cannot be reached.
-
-## 1. Find the endpoint
-
-The default endpoint is:
-
-```text
-http://localhost:5199/api/external-mcp/mcp
-```
-
-Start OpenChatCut first. Open **Settings → MCP** in the trusted editor window,
-copy the displayed bearer-token configuration, and use the endpoint shown there.
-If desktop port 5199 was occupied, that page reflects the active fallback port.
-
-## 2. Register the MCP server
-
-### Codex
-
-Check the existing entry:
-
-```bash
-codex mcp get openchatcut
-```
-
-If it is missing, export the token copied from **Settings → MCP** and register it:
-
-```bash
-export OPENCHATCUT_MCP_TOKEN='<token>'
-codex mcp add openchatcut \
-  --url http://localhost:5199/api/external-mcp/mcp \
-  --bearer-token-env-var OPENCHATCUT_MCP_TOKEN
-```
-
-If an existing URL or token setting is stale, replace only that entry:
-
-```bash
-export OPENCHATCUT_MCP_TOKEN='<token>'
-codex mcp remove openchatcut
-codex mcp add openchatcut \
-  --url http://localhost:5199/api/external-mcp/mcp \
-  --bearer-token-env-var OPENCHATCUT_MCP_TOKEN
-```
-
-### Claude Code
-
-```bash
-claude mcp add --transport http \
-  -H "Authorization: Bearer <token>" \
-  openchatcut http://localhost:5199/api/external-mcp/mcp
-```
-
-For another client, register the endpoint as a Streamable HTTP MCP server and
-set its `Authorization` header to `Bearer <token>`.
-Clients that honor MCP `tools/list_changed` can reduce their initial prompt by
-negotiating progressive exposure:
-
-```text
-http://localhost:5199/api/external-mcp/mcp?toolExposure=progressive
-```
-
-The equivalent request header is
-`X-OpenChatCut-Tool-Exposure: progressive`. The initial list keeps connection,
-session, project-read, `ToolSearch`, and `load_skill` tools. `ToolSearch` and
-`load_skill` expand only that transport session's list. Clients that cache one
-fixed tool list for the whole connection must use the default URL, which retains
-the full compatibility surface.
-
-
-## 3. Verify
-
-When the MCP tools are available, call:
-
-1. `openchatcut_status`
-2. `list_projects`
-
-Interpret the result:
-
-- Editors listed: the live editor bridge is ready.
-- Projects listed but no editor connected: project discovery works; open the
-  intended project using its `editorUrl`.
-- No projects: the connection works; ask whether to create a project.
-- Connection error: follow `known-errors.md`.
-
-Do not create or select a project just because one looks plausible.
-
-## 4. Start the first task
-
-Once the user has identified a project:
-
-1. Call `target_project`.
-2. Surface the URL returned by `get_editor_url`.
-3. Follow `editing-workflow.md`.
-
-## Token lifecycle
-
-The endpoint always requires a bearer token, including on localhost. By default,
-OpenChatCut generates it in private server memory and shows it only through the
-trusted editor's **Settings → MCP** page. A server restart changes that generated
-token, so reconnect clients using the newly displayed configuration.
-
-`OPENCHATCUT_MCP_TOKEN` overrides the generated token when a stable deployment
-secret is required. Keep it in the MCP client's secret/environment configuration;
-never write it into a repository, project document, chat, or browser storage.

package/skills/openchatcut/references/known-errors.md

@@ -1,62 +0,0 @@
-# OpenChatCut MCP recovery
-
-## Connection refused
-
-OpenChatCut is closed, the configured URL is stale, or desktop port 5199 fell
-back to another port.
-
-1. Start OpenChatCut.
-2. Read the endpoint from **Settings → MCP** or the startup log.
-3. Update the single `openchatcut` MCP entry.
-4. Call `openchatcut_status` again.
-
-## Projects exist but no editor is connected
-
-Call `get_editor_url` for the intended project and have the user open it.
-Project listing and creation can work without a live editor, while timeline
-tools require the target project to be open.
-
-## Tool missing
-
-The editor registers project tools after its bridge connects. Open the target
-project, call `openchatcut_status`, and refresh the MCP tool list.
-
-## Active edit session
-
-`begin_edit_session` rejects an existing draft unless `reuseExisting: true` is
-set. Reuse keeps the session ID, staged operations and approval mode. It can
-reuse a draft owned by the same MCP transport or adopt an unchanged orphan.
-It cannot take a draft from another online transport or a recovery in progress,
-and it does not bypass stale-project checks.
-
-## Stale edit session
-
-Call `list_edit_sessions` before starting another session. For an entry marked
-`orphaned: true`, call `recover_edit_session` with one of its `recoveryActions`:
-
-- `resume` continues a draft only when its checkpoint still matches the project.
-- `discard` removes the abandoned draft without changing the live project.
-
-An online owner must finish or discard its own session. Terminal `stale`,
-`cancelled`, and `failed` sessions cannot be resumed; begin a fresh session.
-An auto session stays auto and does not fall back to manual review.
-
-## Proposal awaiting review
-
-The draft is ready, but manual approval is still pending in OpenChatCut. Keep
-polling `get_edit_session` only when the client needs the final state. Report
-`applied`, `rejected`, or `discarded` exactly as returned.
-
-## Skill baseline is newer
-
-Update the installed skill, then re-read its files:
-
-```bash
-npx skills update openchatcut
-```
-
-If the source alias is unavailable:
-
-```bash
-npx skills add 0xsline/OpenChatCut --skill openchatcut
-```

package/skills/openstoryline-install/SKILL.md

@@ -1,228 +0,0 @@
----
-name: openstoryline-install
-description: "Install, configure, and start FireRed-OpenStoryline from source on a local machine. Use when a user asks to set up OpenStoryline, troubleshoot installation, download required resources, fill config.toml API keys, or launch the MCP and web services, as well as Chinese requests like “安装 OpenStoryline”, “配置 OpenStoryline”, “启动 OpenStoryline”, “把 OpenStoryline 跑起来”, “修复 OpenStoryline 安装问题”, or “排查 OpenStoryline 启动失败”. Runs in AICUT: the film is film.html, a page per shot, drawn by `aicut look` and `aicut render`."
-license: Apache-2.0
-source: "FireRedTeam/FireRed-OpenStoryline"
----
-
-# OpenStoryline Install
-
-## 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.
-
-
-Use this skill when the task is to install or repair a local source checkout of FireRed-OpenStoryline.
-
-Keep the workflow deterministic:
-
-1. Confirm the repo path and read the current README.md and config.toml.
-2. Detect local prerequisites before changing anything.
-3. Prefer a local `venv` install unless the user explicitly asks for Docker or `conda`.
-4. Download resources only after Python dependencies succeed.
-5. Validate imports and config loading before claiming success.
-6. This skill assumes macOS, Linux, or WSL with a POSIX shell.
-
-## What This Skill Covers
-
-- Clone the GitHub repo if needed
-- Create a Python environment
-- Install Python dependencies
-- Download `.storyline` models and `resource/` assets
-- Fill `config.toml` model settings
-- Start MCP and web servers
-- Explain common installation/documentation gaps
-
-## Preconditions
-
-Check these first:
-
-- `git`
-- Python `>= 3.11`
-- `ffmpeg`
-- `wget`
-- `unzip`
-
-Optional:
-
-- `docker`
-- `conda`
-
-If `ffmpeg`, `wget`, or `unzip` are missing, install them through the OS package manager before continuing.
-
-Examples:
-
-- macOS with Homebrew:
-  ```bash
-brew install ffmpeg wget unzip
-```
-```
-- Debian/Ubuntu:
-```
-```bash
-sudo apt-get update
-  sudo apt-get install -y ffmpeg wget unzip
-```
-```
-If no supported package manager or permission is available, stop and report the missing system dependency clearly.
-```
-
-## Interpreter selection
-
-First prefer any interpreter that already exists and passes version checks:
-
-1. A system Python `>= 3.11`
-2. An already available conda Python `>= 3.11`
-3. An already available pyenv Python `>= 3.11`, but only if basic stdlib modules work
-
-Validate candidate interpreters before using them:
-
-```bash
-/path/to/python -c "import ssl, sqlite3, venv; print('stdlib_ok')"
-```
-```
-If no supported interpreter already exists, peferr conda fallback:
-```
-```bash
-conda create -y -n openstoryline-py311 python=3.11
-conda run -n openstoryline-py311 python --version
-conda run -n openstoryline-py311 python -m venv .venv
-```
-```
-After a supported interpreter is found, always create a repo-local .venv and continue using .venv/bin/python for install, config validation, and service startup.
-
-Do not duplicate the rest of the workflow for pyenv or conda unless the user explicitly asks to stay inside a conda environment.
-```
-
-## Configuration
-
-Before starting the app, update config.toml.
-
-You can use scripts/update_config.py.
-
-At minimum, fill:
-
-```bash
-.venv/bin/python scripts/update_config.py --config ./config.toml --set llm.model=REPLACE_WITH_REAL_MODEL
-.venv/bin/python scripts/update_config.py --config ./config.toml --set llm.base_url=REPLACE_WITH_REAL_URL
-.venv/bin/python scripts/update_config.py --config ./config.toml --set llm.api_key=sk-REPLACE_WITH_REAL_KEY
-
-.venv/bin/python scripts/update_config.py --config ./config.toml --set vlm.model=REPLACE_WITH_REAL_MODEL
-.venv/bin/python scripts/update_config.py --config ./config.toml --set vlm.base_url=REPLACE_WITH_REAL_URL
-.venv/bin/python scripts/update_config.py --config ./config.toml --set vlm.api_key=sk-REPLACE_WITH_REAL_KEY
-```
-```
-Optional but common:
-
-- `search_media.pexels_api_key` for searching media
-- TTS provider keys under `generate_voiceover.providers.*` (choose one provider)
-```
-
-## Verification
-
-Run these checks before saying installation is complete:
-
-```bash
-.venv/bin/pip check
-PYTHONPATH=src .venv/bin/python -c "from open_storyline.config import load_settings; load_settings('config.toml'); print('config_ok')"
-```
-```
-Also confirm key resources exist:
-```
-```bash
-test -f .storyline/models/transnetv2-pytorch-weights.pth
-test -d resource/bgms
-```
-
-## Start Services
-
-There are two common paths. These are long-running processes. Do not wait for them to exit normally. Treat successful startup log lines or confirmed listening ports as success, and keep the services running in separate shells/sessions as needed.
-
-Manual start:
-
-```bash
-PYTHONPATH=src .venv/bin/python -m open_storyline.mcp.server
-```
-```
-In a second shell:
-```
-```bash
-PYTHONPATH=src .venv/bin/python -m uvicorn agent_fastapi:app --host 127.0.0.1 --port 8005
-```
-
-## Expected Outputs
-
-After a successful install:
-
-- `.venv/` exists
-- MCP listens on the configured local port (commonly `127.0.0.1:8001`)
-- Web listens on the configured web port (commonly `127.0.0.1:8005`, though `run.sh` defaults may differ)
-
-## Common Problems
-
-### `download.sh` is slow or interrupted
-
-Symptom:
-
-- Large downloads stall or reconnect
-
-Fix:
-
-- Let `wget` continue; it supports resume behavior here
-- Verify extracted outputs instead of trusting the progress meter
-
-### Web/MCP server fails to bind
-
-Symptom:
-
-- `operation not permitted` while binding `127.0.0.1` or `0.0.0.0`
-
-Fix:
-
-- In agent sandboxes, request permission to open local listening ports
-- Prefer `127.0.0.1` over `0.0.0.0` unless external access is required
-
-## Response Pattern
-
-When reporting status to the user, separate:
-
-- what is installed
-- what is still downloading
-- what config is still missing
-- what address the service is listening on
-
-Do not say "installation complete" if only the Python packages are installed but the resource bundle is still missing.
-
-## Source
-
-Rewritten for AICUT from [`FireRedTeam/FireRed-OpenStoryline`](https://github.com/FireRedTeam/FireRed-OpenStoryline)
-(Apache-2.0), .claude/skills/openstoryline-install/SKILL.md. The method is that skill's own; the way it runs is AICUT's.
-
-The repository holds 7 skills; this is the one its entry stands for.

package/skills/orca-motion/SKILL.md

@@ -1,100 +0,0 @@
----
-name: orca-motion
-description: "把产品、数据抽象成几种简单图形（格子、横条、卡片、线），做场景内部的 MG 动画：成片元素依次亮起、列表行标色收拢、折线画出、扫描线、镜头推近、卡片弹出、点阵拼字、数字滚动、可变字体标题波浪、打字机。逐帧可录成视频。和 orca-transition-skill 分工：本 skill 管一个场景里面怎么动，那个 skill 管场景和场景之间怎么接。用于「做个产品动效 / MG 动画 / 标题开场动画 / 文字动效 / 苹果发布会那样的产品介绍动画 / 数据可视化动效 / 把这个功能做成动画演示」这类请求。 Runs in AICUT: the film is film.html, a page per shot, drawn by `aicut look` and `aicut render`."
-license: MIT
-source: "zhenwusw/orca-motion-skill"
----
-
-# orca-motion-skill
-
-## 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.
-
-
-把一个产品讲成一段 MG 动画。**成品是若干个场景**，每个场景是一段能逐帧 seek 的时间线；
-场景之间怎么衔接交给 orca-transition-skill。
-
-## 分工
-
-| | orca-motion-skill（本 skill） | orca-transition-skill |
-| --- | --- | --- |
-| 管什么 | 一个场景**里面**：图形怎么生成、按什么顺序动 | 场景**之间**：上一个场景怎么变成下一个 |
-| 写在哪 | `<div class="mo-scene">` + `OrcaMotion.scene(...)` | `<section data-st="…">` |
-| 典型动作 | 格子依次亮起、卡片弹出、拼字、数字滚动、场景内镜头推近 | 放大进元素内部、整页推近、形态变换、遮挡剪辑 |
-
-两个 skill 要装在同一层目录下（比如 `~/.claude/skills/orca-motion-skill` 和 `~/.claude/skills/orca-transition-skill`）。
-
-## 流程（按顺序，不许跳）
-
-0. **装依赖**：本 skill 根目录下没有 `node_modules/` 时，先在根目录跑 `npm install`。
-1. **读原则**：`references/principles.md`。主角是文字本身（标题开场、金句）时，再读 `references/kinetic-type.md`。
-2. **找主角和抽象**：写出这段片子的**一个**主角（一段被空出来的时间、一笔省下的钱），
-   它用什么颜色（主题的 `--accent`，整片只给它用），产品里的东西抽象成哪几种图形（最多 4 种）。
-3. **写场景计划**：每个场景一行：这个场景里主角发生了什么（一个场景只讲一件事）。
-   再写场景之间的转场计划（按 orca-transition-skill 的 `transitions.md` 选手法）。计划写成 HTML 注释放进文件。
-4. **写场景**：读 `references/api.md`，每个场景一段 `OrcaMotion.scene(...)`，数据写在场景里的 JSON。
-   单个场景先用 `templates/scene-starter.html` 单独预览、录制（填好的样子见 `examples/scene-preview/`），动作对了再往下。
-5. **连成片子**：读 `references/integration.md`，把场景放进 orca-transition-skill 的演示稿，一个场景一页。
-6. **录制并看视频**：
-   ```bash
-node <orca-motion-skill>/scripts/capture.mjs <片子>.html -o <片子>.mp4
-```
-```
-输出末尾「页面报告的问题」必须为空。先按正常速度看一遍，再抽帧对照 `principles.md` 末尾的自检标准。
-```
-
-## 本 skill 带了什么
-
-| 路径 | 内容 |
-| --- | --- |
-| `runtime/motion.js` | 场景运行时：`OrcaMotion.scene / build / stage` 和 `m.*` 动作 |
-| `runtime/motion.css` | 场景图形类（`.mo-*`），只用设计变量 |
-| `runtime/tokens.css` `runtime/themes/` | 设计变量和主题，从 orca-transition-skill 同步过来，不要在这里改 |
-| `scripts/capture.mjs` | 逐帧录制，从 orca-transition-skill 同步过来 |
-| `scripts/sync-shared.sh` | 同步上面两项 |
-| `templates/scene-starter.html` | 单独预览一个场景 |
-| `examples/clearing/` | 完整示例：全年热力图 → 推进那一天 → 空档变黄 → 缩回全年拼出 CLEARING |
-| `examples/sieve/` | 完整示例：账单列表 → 订阅标橙收拢成柱子 → 推进看哪些没在用 → 缩回一年的折线 |
-| `examples/scene-preview/` | 单场景预览：模板填好的样子，不需要 orca-transition-skill，浏览器直接打开就播 |
-| `examples/kinetic-title/` | 文字做主角：可变字体标题开场（逐字字重字宽波浪、打字机署名、标志定版后剪到角上），拆解见 `references/kinetic-type.md` |
-
-## 不许做的
-
-- 不许在场景脚本里直接写 `gsap.to` / `gsap.set`、`setTimeout`、CSS 动画。只用 `m.*` 往 `tl` 上排。
-  `m.*` 不够用时，报告缺什么，不要自己绕。
-- 不许写死颜色、字号：颜色只用 `.mo-*` 类和主题变量，文字只用 `.mo-t-*` 五档。
-- 强调色（`--accent`）只给主角用，不许拿来装饰。
-- 不许改 `runtime/` 下的文件。同步来的文件去 orca-transition-skill 改。
-
-## Source
-
-Rewritten for AICUT from [`zhenwusw/orca-motion-skill`](https://github.com/zhenwusw/orca-motion-skill)
-(MIT), SKILL.md. The method is that skill's own; the way it runs is AICUT's.

package/skills/orca-motion/references/api.md

@@ -1,100 +0,0 @@
-# 场景 API
-
-## 写一个场景
-
-```html
-<div class="mo-scene" data-scene="year">
-  <script type="application/json">{ "year": 2027, "events": 2184 }</script>
-</div>
-
-<script>
-  OrcaMotion.scene("year", ({ el, tl, m, data }) => {
-    const g = m.grid({ x: 142, y: 470, rows: 7, cols: 53, cell: 24, gap: 7, levels });
-    const t = m.appear(g.cells, { at: 0.2, origin: "left", span: 1.3 });
-    m.count(stats, { at: 0.3, to: data.events, format: (s) => `${s} events` });
-  });
-</script>
-```
-
-- `el`：场景元素，默认铺满 1920×1080 画布。所有坐标都是画布像素。
-- `tl`：这个场景的时间线。`m.*` 都往它上面排，`at` 是秒。
-- `m`：下面这些动作。大多数返回「这个动作结束的时刻」，方便接着排：`let t = m.appear(...); m.pop(card, t + 0.2)`。
-- `data`：场景里 JSON 的内容。数据写在 JSON 里，脚本里不写死数字。
-- 需要随机数（热力图深浅）用固定种子的伪随机数，保证每次生成一样。
-
-## 生成图形
-
-| 方法 | 作用 |
-| --- | --- |
-| `m.add(cls, {x, y, w, h}, text?, parent?)` | 建一个绝对定位的元素 |
-| `m.backdrop({at?})` | 背景渐变：底色上一团柔光（跟着主题色变），压在场景最底下。`at: [x%, y%]` 光的位置，默认 `[50, 25]` |
-| `m.world()` | 世界层：要被 `m.camera` 推拉的图形放进去（`parent: world`），标题、统计放在外面 |
-| `m.grid({x, y, rows, cols, cell, gap, levels, parent?})` | 格子。`levels[r][c]` 取 0–4 决定深浅。返回 `{cells, at(r, c), box(r, c), width, height}` |
-| `m.agenda({x, y, w, hourH, hours: [开始, 结束], events, parent?})` | 日程时间轴。`events: [{title, start: "13:30", end: "15:00"}]`。返回 `{yOf(时间), blocks, labels}`。放不下两行的短日程自动排成一行 |
-| `m.rows({x, y, w, rowH, gap, columns, items, cls?, parent?})` | 列表，一行一条记录。`columns: [{key, w?, align?, cls?}]`，不写 `w` 的列占剩下的宽度。返回 `{rows, cell(i, key), box(i)}` |
-| `m.line({x, y, w, h, values, min?, max?, parent?})` | 折线，`values` 从左到右均匀分布。返回 `{el, at(i), length}`，用 `m.draw` 画出来 |
-
-## 排动作
-
-| 方法 | 作用 |
-| --- | --- |
-| `m.appear(els, {at, origin, span, kind, duration})` | 成片依次出现。`origin`：`left` `right` `top` `center` 或 `[x, y]`；`kind`：`pop` `rise` `fade` |
-| `m.pop(el, at)` `m.rise(el, at)` `m.fade(el, at)` `m.hide(el, at)` | 单个元素弹出 / 上浮 / 淡入 / 淡出 |
-| `m.reveal(el, at, duration)` | 文字从左到右揭开 |
-| `m.count(el, {at, from, to, duration, decimals, ease, format})` | 数字滚动，`format(已格式化的数字)` 返回要显示的整句 |
-| `m.swapText(el, text, at)` | 换一句文字（出去再进来） |
-| `m.heat(cell, at)` | 一个格子变成强调色并弹一下 |
-| `m.mark(el 或 [el], at, {cls, on, pulse})` | 任意元素加上 / 去掉一个类（默认 `mo-hot`），轻轻弹一下。`on: false` 去掉（不再是主角） |
-| `m.dim(el, at, {to})` | 调暗成背景信息（默认 0.3），不是隐藏 |
-| `m.move(el, {x, y, w, h}, {at, duration})` | 移动 / 变形到新位置和尺寸，缺的项不变。同一个元素可以连续 move（列表行收拢成柱子） |
-| `m.turn(el, deg, {at, duration, origin})` | 旋转到 `deg` 度（顺时针为正），同一个元素可以连续 turn。`origin` 是旋转中心，默认 `"50% 50%"`；扇形展开绕底边中点转用 `"50% 100%"` |
-| `m.draw(line, {at, duration, follow})` | 折线从起点画到终点。`follow`：跟着笔尖走的元素（圆点），初始位置放在起点 |
-| `m.spell(grid, "TEXT", {at, row, col, each})` | 在网格里点亮格子拼出 5×7 点阵字（A–Z、0–9），每个字占 5 列，字间空 1 列 |
-| `m.scan(el, {at, from, to, duration})` | 扫描线从 y=from 走到 y=to |
-| `m.camera(world, target, {at, duration, fill})` | 场景内镜头：把 `target`（元素或 `{x, y, w, h}`）推到画面中间，宽度占画面 `fill` |
-| `m.blur(el 或 [el], px, {at?, duration?})` | 景深：退到焦外。不写 `at` 是静态的（第 0 秒已经虚了）；写 `at` 在时间线上过渡，`px = 0` 回到清晰 |
-| `m.cut(el 或 [el], at, {on})` | 剪切：`at` 瞬间出现（`on: false` 瞬间消失），不是淡入淡出 |
-| `m.settle(el, at, {from, to, duration, ease})` | 定版：从 `from` 倍缩到 `to` 倍，减速停住（标志、大标题入场） |
-| `m.hold(t)` | 让时间线至少持续到 `t` 秒（结尾停住），代替 `tl.to({}, …)` |
-| `m.flip(at, (on) => …)` | 其他瞬间切换。`on` 为 false / true 分别是 at 之前 / 之后的状态，两种状态都要写 |
-
-## 文字
-
-文字做主角时（可变字体标题、打字机），原则和案例见 `kinetic-type.md`。
-
-| 方法 | 作用 |
-| --- | --- |
-| `m.glyphs(el)` | 把元素里的文字拆成一个字一个 `<span>`，返回这些 span |
-| `m.vary(glyphs, {at, span, keys, ease})` | 可变字体轴波浪：每个字按 `keys`（`[{t, wght, wdth, …}]`，`t` 相对这个字的时刻）走一遍字体轴，第 i 个字晚 `i/(n−1) × span` 秒 |
-| `m.type(el, text, {at, duration})` | 打字机 |
-
-## 图形类（`motion.css`）
-
-| 类 | 样子 |
-| --- | --- |
-| `.mo-t-xl` 120 / `.mo-t-l` 64 / `.mo-t-m` 40 / `.mo-t-s` 24 / `.mo-t-xs` 16 等宽 | 文字只用这五档 |
-| `.mo-dim` `.mo-strong` `.mo-on-accent` | 文字颜色：暗 / 亮 / 强调色底上的字 |
-| `.mo-mono` `.mo-regular` | 换成等宽字 / 常规字重（和 `.mo-t-*` 一起用） |
-| `.mo-cell` `.mo-l0`–`.mo-l4` | 格子，深浅五级 |
-| `.mo-hot` | 主角（强调色）：格子、列表行都能加 |
-| `.mo-row` `.mo-row-cell` | 列表行和单元格（`m.rows` 生成） |
-| `.mo-line` `.mo-axis` | 折线（强调色）、坐标轴 |
-| `.mo-panel` | 大面板 |
-| `.mo-event` | 日程块（`m.agenda` 生成） |
-| `.mo-rule` `.mo-dot` | 主角的线和点（强调色） |
-| `.mo-dash` | 虚线框：标出一块区域 |
-| `.mo-card` | 主角卡片（强调色） |
-
-## 给转场留锚点
-
-场景里生成的元素写 `el.dataset.zoom = "键"`，orca-transition-skill 的 `zoom-into` 就能用它：
-
-```js
-grid.at(3, 51).dataset.zoom = "day";
-```
-
-## 确定性
-
-- 所有动作都在 `tl` 上。不许用 `setTimeout`、`requestAnimationFrame`、CSS 动画、`Date.now()`。
-- 瞬间切换（换文字、加类名）用 `m.flip` / `m.swapText` / `m.heat`，不要用 `tl.call`：
-  `call` 只在往前播时触发，录制和预览时往回拖，状态不会复原。

package/skills/orca-motion/references/integration.md

@@ -1,59 +0,0 @@
-# 连成片子：放进 orca-transition-skill 的演示稿
-
-一个场景一页。片子本身就是一份 orca-transition-skill 的演示稿，照它的规范写转场计划和转场。
-
-## 文件结构
-
-```html
-<head>
-  <link rel="stylesheet" href="{{T}}/node_modules/reveal.js/dist/reset.css">
-  <link rel="stylesheet" href="{{T}}/node_modules/reveal.js/dist/reveal.css">
-  <link rel="stylesheet" href="{{T}}/runtime/tokens.css">
-  <link rel="stylesheet" href="{{T}}/runtime/themes/clearing.css">   <!-- 主题 -->
-  <link rel="stylesheet" href="{{M}}/runtime/motion.css">
-</head>
-<body>
-<div class="reveal">
-<!-- 场景计划 + 转场计划 -->
-<div class="slides">
-  <section data-hold="1">
-    <div class="mo-scene" data-scene="year"><script type="application/json">{…}</script></div>
-  </section>
-  <section data-st="zoom-into" data-st-direction="in" data-st-from="day" data-hold="1">
-    <div class="mo-scene" data-scene="day"><script type="application/json">{…}</script></div>
-  </section>
-</div>
-</div>
-<script src="{{T}}/node_modules/reveal.js/dist/reveal.js"></script>
-<script src="{{T}}/node_modules/gsap/dist/gsap.min.js"></script>
-<script src="{{M}}/runtime/motion.js"></script>
-<script> OrcaMotion.scene("year", …); OrcaMotion.scene("day", …); </script>
-<script src="{{T}}/runtime/engine.js"></script>
-</body>
-```
-
-- `{{T}}` = 到 orca-transition-skill 根目录的相对路径，`{{M}}` = 到本 skill 根目录的相对路径。
-- **脚本顺序不能变**：reveal → gsap → motion.js → 场景注册 → engine.js。
-  engine.js 一加载就会开始第一页，场景要在它之前注册好。
-- gsap 只加载一份（用 orca-transition-skill 的）。
-
-## 场景什么时候播
-
-- 翻到这一页：**先把场景图形建好停在第 0 秒**，再跑转场，转场结束后场景从头播。
-  所以转场看到的新页，是场景第 0 秒的样子。用 `zoom-into` / 匹配放大推进来的场景，第 0 秒就要有完整内容，
-  不要让图形在场景里才出现（见 `principles.md` 第 4 节）。
-- 往回翻、跳页：没有转场，场景直接从头播。
-- 离开这一页：场景直接跳到结尾，下一次转场拍到的是场景的最终画面。
-- 录制时：停留时长 = max(`data-hold`, 场景时长 + 0.8 秒)，停留期间逐帧录场景。
-
-## 场景之间用什么转场
-
-| 两个场景的关系 | 转场 |
-| --- | --- |
-| 下一个场景是上一个场景里某个元素的细节（格子 → 那一天） | `zoom-into` in，`data-st-from` 指上一个场景里写了 `data-zoom` 的元素 |
-| 下一个场景是上一个场景的全局（那一天 → 全年） | `zoom-into` out，`data-st-to` 指这个场景里写了 `data-zoom` 的元素 |
-| 同一批图形换个排法 | 形态变换 / 多合一 / 一变多（写法见 orca-transition-skill 的 `transitions.md`） |
-| 换话题 | 遮挡剪辑 |
-
-`zoom-into` 的锚点是场景生成出来的元素，所以一定要在场景脚本里写 `dataset.zoom`。
-out 方向的锚点最好**已经是主角颜色**（`.mo-hot`）：观众看到「那一天缩回来变成了一个黄格子」。