用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/tools-only/X-Skills --skill yt-pipeline命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | yt-pipeline |
| description | Run the full YouTube content production pipeline |
| argument-hint | ["topic-or-channel-urls"] |
| allowed-tools | Read, Write, Edit, Bash, Grep, Glob, Task, WebSearch, AskUserQuestion |
| model | sonnet |
Run the complete YouTube content production pipeline. This is a five-phase workflow that takes a topic or niche focus and produces a fully developed video outline ready for demo prep and filming.
You are the orchestrator. Your job is coordination, sub-agent spawning, data merging, and user communication. Never do batch processing yourself - delegate ALL batch work to sub-agents.
Before doing anything, collect inputs from the user using AskUserQuestion:
Niche/Tool Focus: What tool, feature, or topic area are we researching? (e.g., "AI writing tools for professionals", "new software features", "productivity automation")
Competitor Channels (optional): Any specific YouTube channels to analyze?
Specific Feature/Update (optional): Is there a specific feature drop or update to cover? If so, this is likely an Update Video with fast turnaround.
Do NOT proceed until the focus is confirmed.
Use web research and optional sub-agents for parallel execution:
yt-scraper agent for YouTube channel data collection (if Apify MCP available)channel-analyzer agents in batches of 3 channels each for competitive analysisAfter all research completes, merge results into:
niche-analysis.json - structured dataniche-report.md - human-readable summaryReport to user:
Research complete.
- Channels analyzed: [N]
- Videos scraped: [N]
- Content gaps identified: [N]
- Outlier videos found: [N]
Present the niche report and ask: "Here's what I found. Ready to move to ideation, or want to research more channels?"
This is a mandatory human checkpoint. Do NOT proceed without approval.
Using research findings, generate 15-20 video ideas. Each idea must include:
Prioritize Tier 1 ideas. Include a mix of content types.
Spawn idea-validator sub-agents in a single message. Each validates a batch of 5 ideas.
Merge validation results and present a ranked list:
Here are the validated ideas, ranked by opportunity score:
[Table: Rank | Title | Tier | Type | Search Demand | Competition | Score]
Which ideas do you want to develop into briefs?
This is a mandatory human checkpoint.
For each idea the user selects:
skills/yt-brief/SKILL.mdvideo-brief-{slug}.md"Here's the brief for '{idea title}'. Review and approve?"
This is a mandatory human checkpoint.
For each approved brief:
skills/yt-packaging/SKILL.mdpackaging-{slug}.md"Here are the title and thumbnail options. Pick your favorites."
This is a mandatory human checkpoint.
For each packaged video:
skills/yt-outline/SKILL.mdvideo-outline-{slug}.md + demo-prep-checklist-{slug}.md"Here's the outline and demo prep checklist. Ready for the team to prep and film."
Pipeline Complete.
- Ideas generated: [N]
- Ideas validated: [N]
- Briefs created: [N]
- Videos fully packaged and outlined: [N]
- Output files: [list]
| Agent | subagent_type |
|---|---|
| YouTube Scraper | youtube-strategy:yt-scraper |
| Channel Analyzer | youtube-strategy:channel-analyzer |
| Idea Validator | youtube-strategy:idea-validator |
ALL sub-agents in a phase MUST be spawned in ONE message using the Task tool. This is the only way to trigger parallel execution.
基于 SOC 职业分类