Short video publish experiment
Agent Skills workflow for Chinese talking-head video operations, scripting, production, publishing preparation, and review.
npx -y skills add slalomboy/talking-head-video-suite --skill short-video-publish-experimentAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 22 days oldThe repository was created 22 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
- 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its author says it does
Copied from the file, not written here
Use when reviewed short-video content needs a publish hypothesis, controlled variable, comparable conditions, metrics, stop rules, or an explicit release approval gate.
SKILL.md
1.3 KB, 369 tokens by cl100k_base, as published. Nobody here has run it
Short Video Publish Experiment
核心原则
平台反馈是实验数据，不是玄学判决；单条结果只能形成观察，不能单独证明因果。
触发与边界
用于自然发布实验设计、版本对比、指标与停止条件。发布属于外部行动，本 Skill 只准备计划，默认 pendingApproval，不执行发布。
方法
- 写清
hypothesis和要回答的问题。 - 一轮只指定一个
primaryVariable；其他内容形成comparableConditions。 - 选择与任务匹配的
metric，同时记录受众和承接质量。 - 预先写
stopRule、观察窗口、样本要求和版本标识。 - 行动前核验当前平台官方规则、内容审查和明确授权。
输出契约
只写 publishExperiment：hypothesis、primaryVariable、comparableConditions、metric、stopRule、规则核验、版本和 pendingApproval。未授权时不得写成已发布。
常见错误
用两条不可比视频推断因果；同时改开篇、场景、时长和受众；只看播放；先发布再补授权；把规则经验写成固定事实。
What ships with it: 3 files
498 B alongside SKILL.md
agents/
- openai.yaml276 B
assets/
- template.md83 B
references/
- method.md139 B