youtube-research-skills
AGENTS.md91 starsChanged 11 days agoMITView on GitHub
This repository is a set of Agent Skills for YouTube research on top of the TubeLab MCP. Skills live in skills/<name>/SKILL.md. Read the one that matches the ask, then follow it. skills/tubelab-mcp is the data layer every workflow skill names; load it when a tool call fails, returns nothing, or you are unsure which filter to use. Ordered chains for common jobs live in playbooks/ (run the whole agency, launch a faceless channel, monthly client report). The plugin ships six…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,439 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository18 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- audit-channel-riskSkill
- channel-analysisSkill
- comment-miningSkill
- competitor-researchSkill
- monetization-researchSkill
- niche-bendingSkill
- niche-finderSkill
- niche-validatorSkill
- niche-watchlistSkill
- reverse-engineer-channelSkill
- script-breakdownSkill
- sponsorship-prospectingSkill
- thumbnail-conceptsSkill
- title-formulasSkill
- trend-reportSkill
- tubelab-mcpSkill
- video-ideationSkill
The file
We split this AGENTS.md into the rules it gives an agent
5 rules. About 0 are common, 4 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 44
Always set publishedAtFrom (tokens: month, trimester, semester, year).
- Distinctive: line 47
Every finding carries its proof from a tool response: channel (handle) · video title verbatim · views · ratio · date.
- Distinctive: line 49
Look at the thumbnail of every video you analyze.
- Distinctive: line 52
Write reports in plain language.
- Distinctive: line 62
No MCP: the REST reference at https://tubelab.net/docs/skill.md.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Agent routing guide
This repository is a set of Agent Skills for YouTube research on top of the TubeLab MCP. Skills live in skills/<name>/SKILL.md. Read the one that matches the ask, then follow it. skills/tubelab-mcp is the data layer every workflow skill names; load it when a tool call fails, returns nothing, or you are unsure which filter to use.
Which skill for which ask
| The user asks | Load |
| Find a niche, unsaturated niches, faceless ideas, niches that earn beyond AdSense, twist or combine niches | niche-finder |
| Bend this channel's format into another market, niche bending, which proven formats can I bring to my market | niche-bending |
| What is working in X, why did this video blow up, trend report, weekly digest, hooks and formats to steal | trend-report |
| Video ideas for my channel, what should I make next | video-ideation |
| Analyze or roast my channel, why did views drop, what to change first | channel-analysis |
| Will my channel get demonetized or terminated, is my channel at risk, audit my channels for monetization risk | audit-channel-risk |
| Reverse engineer a channel, what works for it, its title, thumbnail and hook formulas, how it grew and earns, what can I copy from it | reverse-engineer-channel |
| Thumbnail for this video or title or script, thumbnail concepts, generate thumbnails, which thumbnails work in a niche | thumbnail-concepts |
| Is X worth entering, is this niche saturated, go or no-go, score this niche, compare a shortlist of niches | niche-validator |
| Titles for my video, title formulas, trending title templates, rewrite a title, a title and a thumbnail together (titles first) | title-formulas |
| Who are my competitors, how do I compare, competitive landscape, who is growing in my space | competitor-research |
| What do viewers ask or complain about, voice of audience, comments | comment-mining |
| Script structure, beat sheet, hook of this video, script skeleton, how the top videos of a niche or channel are built | script-breakdown |
| Monetization gaps for my channel, what to sell, how channels like mine make money, raise RPM, who could sponsor my channel (the creator's side) | monetization-research |
| Which channels should we sponsor, creator prospect list for a product, match creators to a buyer (the brand's side) | sponsorship-prospecting |
| Watch a niche, which channels broke out since last run, scheduled digest | niche-watchlist |
| How do TubeLab tools work, which filter, credits, REST | tubelab-mcp |
Ordered chains for common jobs live in playbooks/ (run the whole agency, launch a faceless channel, monthly client report).
Roles (Claude Code agents)
The plugin ships six agents in agents/, one per role. Each loads the skills below and returns the skill's report. On other clients, load the skills directly.
| Agent | Skills |
🔭 niche-scout | niche-finder; niche-bending to bend one channel's format, or a proven one, into new markets; niche-watchlist to watch a niche over time |
⚖️ validator | niche-validator |
📈 trend-analyst | trend-report |
🎬 script-analyst | script-breakdown |
🎨 packaging-team | title-formulas, then thumbnail-concepts for the chosen title |
💰 monetization-team | monetization-research; sponsorship-prospecting when the user is a brand choosing channels to sponsor |
Rules every skill follows
- Always set
publishedAtFrom(tokens:month,trimester,semester,year). Widen one step at a time and say so. - Content Authenticity is never off:
classificationQualityat"neutral"or above for niches, outliers, ideation and titles;"positive"for anything thumbnail-related. Never drop it to widen a thin result.niche-watchlistandaudit-channel-riskare the exceptions: the watchlist's saved search and the audit's similar channel searches leave the filter out. - Rank by
averageViewsRatio, not raw views. Show both. - Every finding carries its proof from a tool response: channel (handle) · video title verbatim · views · ratio · date. Invent nothing.
- Link everything you cite: videos, channels, moments, brands and products. Use a URL from tool data, never a guessed one. A sponsor that was only heard links to the video moment where it was said.
- Look at the thumbnail of every video you analyze. Fetch the image from its link and open it.
- Revenue, RPM and audience are estimates. Present as bands, labelled.
- Searches cost credits. Channel, video, transcript and comment reads are free. Search to find candidates, and read to understand them. Most skills use two to six searches, and do not page through a paid search. One skill goes deeper, because the sample is the product:
niche-finderpages its two channel searches 3 to 4 pages deep. - Write reports in plain language. Use short, full sentences, not dense or telegraphic notes, and no jargon. A reader who is not a YouTube strategist should understand every line on first read.
- End with the artifact laid out in the skill's "Output Format" section.
Connecting the MCP
- Claude Code: this repo is a plugin;
.mcp.jsonregistershttps://public-api.tubelab.net/mcpand the first call opens OAuth. Orclaude mcp add --transport http tubelab https://public-api.tubelab.net/mcp. - Claude.ai, Claude Desktop: the TubeLab connector (https://claude.ai/directory/tubelab), OAuth.
- ChatGPT and Codex: the TubeLab app (https://chatgpt.com/plugins/plugin_asdk_app_6ab3980fb1488191a9b26b3f59efd90b), OAuth.
- Grok Bot: ask the Bot to add the MCP server by URL, then paste or point it at a
SKILL.mdand ask it to save it as a skill; Grok Build:/mcpsmodal, skills in.grok/skills/or~/.grok/skills/. - Cursor, others: API key as
Authorization: Bearer <key>(Pro plan). Setup: https://tubelab.net/docs/api/mcp-server - No MCP: the REST reference at https://tubelab.net/docs/skill.md.
Repo conventions for contributors
See CONTRIBUTING.md. Each workflow skill is one self-contained SKILL.md with the sections Overview · Workflow · Output Format · Common Pitfalls. No skill has a references/ folder; the full API reference lives at https://tubelab.net/docs/skill.md. Keep it simple, do not over-engineer, Conventional Commits (type(scope): subject). Run node scripts/validate-skills.mjs before opening a PR.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

