ReelOnce-skill

ReelOnce 一体化总控 skill。单次调用即可完成从输入文本到最终视频输出的完整流程：planning、资产图/分镜图/TTS 生成、镜头视频生成、Remotion 工程生成与最终 MP4 渲染。

Security Scan

The skill's description matches a local text-to-video pipeline, but the SKILL.md expects and uses environment variables and external video services that are not declared in the registry metadata — this mismatch is suspicious and worth caution before installing or running.

Purpose & Capability

The skill is presented as a local orchestrator for a ReelOnce pipeline and the included scripts simply call an installed 'reelonce' package, which is coherent. However, the documentation repeatedly requires/mentions SKILLS_VIDEO_* / CUSTOM_* / COMMERCIAL_* environment variables and instructs users to copy and source env.local — yet the registry metadata lists no required environment variables. That omission is an incoherence: the skill will practically need API keys and environment config to function, but these are not declared.

Instruction Scope

SKILL.md instructs creating a venv, running pip install -e '.[dev]', copying and sourcing env.local, running npm install inside generated Remotion projects, and sending asset references to commercial video services (either URLs or base64-encoded images). These operations are within the stated purpose, but they include reading local files and potentially converting and sending local images as base64 to external endpoints — behavior that should have been made explicit in the metadata and permission model.

Install Mechanism

There is no registry install spec (instruction-only skill with small helper scripts). The README instructs standard pip editable install and npm install in the generated Remotion project. Those are expected for this kind of project and are not themselves suspicious, but they do mean code from the 'reelonce' package and npm packages will be installed and executed on the host.

Credentials

Although the registry lists no required env vars, SKILL.md clearly expects variables like SKILLS_VIDEO_API_KEY, SKILLS_VIDEO_IMAGE_MODEL, SKILLS_VIDEO_VIDEO_MODEL and supports CUSTOM_* / COMMERCIAL_* variants. Requiring API keys and other secrets is proportionate to the task, but the metadata failing to declare them is a mismatch. Also, instructing users to 'set -a && source ./env.local' will export all variables from env.local into the environment — potentially exposing many secrets if env.local contains unrelated credentials.

Persistence & Privilege

The skill is not forced-always (always: false) and does not request system-wide changes in the provided files. It runs subprocesses (calls into the installed 'reelonce' package and will run npm in generated project directories) but does not itself modify other skills or global agent config.

What to consider before installing

Before installing or running this skill: (1) review env.local.example to see which API keys and variables it expects (SKILLS_VIDEO_*, CUSTOM_*, etc.) — the registry metadata omits these; (2) understand that running the skill will install and run the 'reelonce' Python package and npm packages (inspect their provenance and code if you don't trust the source); (3) the pipeline may upload images (URLs or base64) to external video services — don't run it with sensitive local images or credentials present; (4) avoid blindly sourcing env.local from untrusted sources (it can export many secrets); (5) if you want to proceed, run it in an isolated environment (container or VM), inspect the installed reelonce package first, and only provide API keys for services you trust.

reelonce-skill - 一体化文本到视频

功能简介

reelonce-skill 是 ReelOnce 的总控入口。

默认顺序：

1. planning：输入文本，统一完成分镜、角色/场景/道具、声音映射，并写出 narration.json / assets.json
2. asset_generation：生成资产参考图、分镜图、TTS 音频
3. shot_render：生成镜头视频片段
4. remotion_preview --mode react：生成 Remotion React 项目
5. 在 Remotion 工程目录中执行 npm install / npm run render：渲染最终 MP4

安装依赖

在仓库根目录执行以下步骤：

python -m venv .venv
source .venv/bin/activate
pip install -U pip
pip install -e ".[dev]"
cp env.local.example env.local
set -a
source ./env.local
set +a

额外前置依赖：

• Python 3.10+
• Node.js 18+
• npm

说明：

• pip install -e ".[dev]" 会安装 ReelOnce 的 Python 依赖与测试依赖
• 首次执行最终渲染时，skill 会在生成的 Remotion 工程目录内自动执行 npm install
• 当前核心 Python 包导入路径就是 reelonce
• 如果用户还没安装依赖，先明确指引用户在仓库根目录按上面的顺序执行，不要跳过 cp env.local.example env.local 和 source ./env.local
• 如果执行 skill 时出现 无法导入已安装的 ReelOnce Python 包、ModuleNotFoundError、缺少 npm / node 等错误，优先判断为环境未就绪，并按上面的命令重新指引用户安装依赖

执行前检查与提示规则

在真正运行 reelonce-skill 之前，先检查依赖和环境变量是否就绪。

• 如果发现用户尚未安装 Python / Node.js 依赖，先让用户完成“安装依赖”章节中的命令，再继续执行
• 如果发现缺少 env.local，先提示用户从 env.local.example 复制一份，并加载到当前 shell
• 文生图 / 视频生成默认优先推荐 skills.video
• 如果用户没有配置文生图能力，或没有配置视频生成所需的 skills.video 参数，明确提示用户访问 https://skills.video/ 注册并获取 API Key
• 发现以下任一情况时，都按上条提示用户补齐配置：
  • SKILLS_VIDEO_API_KEY 为空
  • SKILLS_VIDEO_IMAGE_MODEL 为空，导致文生图没有可用模型
  • SKILLS_VIDEO_VIDEO_MODEL 为空，导致视频生成没有可用模型
  • 用户既没有配置 SKILLS_VIDEO_*，也没有配置可替代的 CUSTOM_IMAGE_* / CUSTOM_VIDEO_*
• 提示用户时要直接给出可执行修改方式，至少包含：
  • cp env.local.example env.local
  • 在 env.local 中填写 SKILLS_VIDEO_API_KEY=...
  • 按需填写 SKILLS_VIDEO_IMAGE_MODEL 与 SKILLS_VIDEO_VIDEO_MODEL
  • set -a && source ./env.local && set +a

输入

• 故事文本（命令行参数、--input-file 或标准输入）
• 项目 ID（可选，不传则自动生成）
• 语言、分镜数量、图像/TTS/视频模型参数
• 视频参考图来源（shot 分镜图或 asset 资产图）

输出

• output/{project_id}/reelonce/
  • narration.json
  • assets.json
  • pipeline-result.json
• output/{project_id}/animation.db
  • 当前项目独立数据库
• output/{project_id}/remotion/
  • Remotion React 工程
• 最终视频：
  • 默认 output/{project_id}/remotion/final_video.mp4
  • 或使用 --output-file 指定

使用示例

# 最简单的完整流程
/reelonce-skill "一个少年误入废弃游乐园，在雨夜里发现了会说话的木马。"

# 指定项目 ID 和镜头数
/reelonce-skill "故事文本" --project-id demo-001 --shots 8

# 使用资产图作为视频参考图
/reelonce-skill "故事文本" --project-id demo-001 --reference-image-source asset

# 导出每个镜头最终提交给视频服务的 prompt
/reelonce-skill "故事文本" --project-id demo-001 --debug-video-prompt

# 只生成到 Remotion 工程，不渲染 MP4
/reelonce-skill "故事文本" --project-id demo-001 --no-render

# 从文件读取文本，并输出机器可读 JSON
/reelonce-skill --input-file story.txt --json

关键参数

• --project-id：项目 ID，不传则自动生成
• --shots：限定分镜数量；不传则由模型根据故事内容自行决定
• --reference-image-source：视频阶段参考图来源，shot 为分镜图，asset 为资产图
• --debug-video-prompt：把每个镜头最终提交给视频服务的 prompt 保存到 output/{project_id}/render/prompts/
• --output-file：最终视频文件名或路径
• --no-render：只生成 Remotion 工程，不输出最终 MP4
• --no-subtitle：生成 Remotion 工程时不带字幕
• --allow-partial：允许 asset-gen / shot-render 返回 partial 后继续
• --json：仅向 stdout 输出最终结果 JSON，过程日志写到 stderr

注意事项

• 最终 MP4 渲染依赖 Node.js、npm 和 Remotion 相关依赖
• 若首次渲染，skill 会在生成的 Remotion 工程目录中执行 npm install
• asset_generation 和 shot_render 依赖 planning 已经把分镜和资产数据写入本地数据库
• 当前主流程推荐按 env.local.example 使用 CUSTOM_* 环境变量；运行时仍兼容旧 COMMERCIAL_* 变量名
• 商业视频服务的参考图请求体只会发送 URL 或 base64：如果上游已有云存储 URL 就直接传 URL，没有云存储时会把本地图像转成 base64
• 本 skill 自身不重写底层业务逻辑，而是调用 reelonce 包内现有 pipeline
• narration、asset-extractor 已不再作为独立顶层流程暴露；它们对应的能力已收敛到 planning

工作流程位置

[reelonce-skill] = 01-planning -> 02-asset_generation -> 03-shot_render -> 04-remotion_preview -> final mp4