Files
last30days-skill/docs/comparison-results/json/base-1-claude-code.json
T
Matt Van Horn bed0557b65 feat(quality): GOAT synthesis improvements - hybrid cross-source linking, YouTube synonyms, human-readable xref tags
Ran 15-way blinded comparison (5 topics x 3 versions). CROSS won all 5 topics
(4.74/5.0 avg vs HN 4.10, Base 3.73). Then improved CROSS further:

- dedupe.py: hybrid similarity (token+trigram Jaccard) at 0.40 threshold,
  cross-source links went from 3 to 13 items across 5 topics
- render.py: [xref: HN5, HN4] -> [also on: HN, Reddit] for human-readable tags
- youtube_yt.py: SYNONYMS dict so "hip hop" matches "rap" (0.33 -> 0.71 score)
- SKILL.md: instruction #7 tells Claude to lead with cross-platform signals

Validation: improved CROSS scores 4.38/5.0 vs original 3.98 (+0.40), wins 4/5
topics. Biggest gains in specificity (+0.8) and format compliance (+1.0).

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
2026-02-25 16:06:53 -08:00

519 lines
26 KiB
JSON

{
"topic": "Claude Code skills and MCP servers",
"range": {
"from": "2026-01-26",
"to": "2026-02-25"
},
"generated_at": "2026-02-25T20:28:50.287295+00:00",
"mode": "both",
"openai_model_used": "gpt-5.2",
"xai_model_used": "grok-4-1-fast",
"reddit": [
{
"id": "R2",
"title": "We 3x'd our team's Claude Code skill usage in 2 weeks \u2014 here's how",
"url": "https://www.reddit.com/r/ClaudeCode/comments/1rbr5t7/we_3xd_our_teams_claude_code_skill_usage_in_2/",
"subreddit": "ClaudeCode",
"date": "2026-02-22",
"date_confidence": "high",
"engagement": null,
"top_comments": [],
"comment_insights": [],
"relevance": 0.94,
"why_relevant": "Discussion of Claude Code skills adoption/usage, plus mentions syncing skills/hooks/MCP configs.",
"subs": {
"relevance": 94,
"recency": 90,
"engagement": 50
},
"score": 76
},
{
"id": "R4",
"title": "ClaudeInOne \u2014 a full framework for Claude Code, installed in one command",
"url": "https://www.reddit.com/r/ClaudeCode/comments/1rcbria/claudeinone_a_full_framework_for_claude_code/",
"subreddit": "ClaudeCode",
"date": "2026-02-23",
"date_confidence": "high",
"engagement": null,
"top_comments": [],
"comment_insights": [],
"relevance": 0.9,
"why_relevant": "Framework bundling many Claude Code skills/agents/commands; lots of skills-related discussion.",
"subs": {
"relevance": 90,
"recency": 93,
"engagement": 50
},
"score": 75
},
{
"id": "R1",
"title": "Self-improvement Loop: My favorite Claude Code Skill",
"url": "https://www.reddit.com/r/ClaudeCode/comments/1r89084/selfimprovement_loop_my_favorite_claude_code_skill/",
"subreddit": "ClaudeCode",
"date": "2026-02-18",
"date_confidence": "high",
"engagement": null,
"top_comments": [],
"comment_insights": [],
"relevance": 0.96,
"why_relevant": "Directly about a Claude Code Skill (custom skill, how it works, SKILL.md).",
"subs": {
"relevance": 96,
"recency": 76,
"engagement": 50
},
"score": 74
},
{
"id": "R3",
"title": "I built 12 SEO skills for Claude Code (open source)",
"url": "https://www.reddit.com/r/ClaudeCode/comments/1r0l549/i_built_12_seo_skills_for_claude_code_open_source/",
"subreddit": "ClaudeCode",
"date": "2026-02-10",
"date_confidence": "high",
"engagement": null,
"top_comments": [],
"comment_insights": [],
"relevance": 0.93,
"why_relevant": "Concrete example of building and sharing Claude Code skills.",
"subs": {
"relevance": 93,
"recency": 50,
"engagement": 50
},
"score": 66
}
],
"x": [
{
"id": "X6",
"text": "@adamwathan Codex has been crushing it for me with implicit skill usage to the point I was shocked \n\nClaude Code I resorted to explicit mentions and I\u2019ve been pleasantly surprised by how well Codex has been calling out to skills on its own",
"url": "https://x.com/zeeg/status/2026745680195367091",
"author_handle": "zeeg",
"date": "2026-02-25",
"date_confidence": "high",
"engagement": {
"likes": 4,
"reposts": 1,
"replies": 2
},
"relevance": 0.7,
"why_relevant": "",
"subs": {
"relevance": 70,
"recency": 100,
"engagement": 100
},
"score": 86
},
{
"id": "X4",
"text": "\ud83d\udea8 Anthropic just open-sourced the exact Skills library their own engineers use internally.\n\nStop building Claude workflows from scratch.\n\nThese are plug-and-play components that work across Claude Code, API, SDK, and VS Code copy once, deploy everywhere.\n\nWhat's inside:\n\n\u2192 Excel + PowerPoint generation out of the box\n\u2192 File handling and document workflows\n\u2192 MCP-ready subagent building blocks\n\u2192 Pre-built patterns for multi-step automation\n\u2192 Production templates you'd normally spend weeks writing\n",
"url": "https://x.com/ihtesham2005/status/2026752089473314975",
"author_handle": "ihtesham2005",
"date": "2026-02-25",
"date_confidence": "high",
"engagement": {
"likes": 6
},
"relevance": 0.7,
"why_relevant": "",
"subs": {
"relevance": 70,
"recency": 100,
"engagement": 81
},
"score": 80
},
{
"id": "X9",
"text": "Day 7 building my first B2C app\n\nBeen setting everything up this week. Manually built 5 agents to help me in the early stages and set up the Claude Code skills i'll need for the build.\n\nAlso been testing OpenClaw these days, already delegated my expense tracking to it.\n\nNow the scary part: Figuring out the visual identity of the app. my design skills are basically zero, i always used to delegate this to someone with actual talent haha. \n\nAny recommendations to learn the fundamentals? would reall",
"url": "https://x.com/DevTenta/status/2026742028730527786",
"author_handle": "DevTenta",
"date": "2026-02-25",
"date_confidence": "high",
"engagement": {
"likes": 4
},
"relevance": 0.7,
"why_relevant": "",
"subs": {
"relevance": 70,
"recency": 100,
"engagement": 59
},
"score": 74
},
{
"id": "X8",
"text": "@Jompiras Claude code + skills. Muy breve.",
"url": "https://x.com/JorgeJaramillo/status/2026743135435362649",
"author_handle": "JorgeJaramillo",
"date": "2026-02-25",
"date_confidence": "high",
"engagement": null,
"relevance": 0.7,
"why_relevant": "",
"subs": {
"relevance": 70,
"recency": 100,
"engagement": 35
},
"score": 64
},
{
"id": "X5",
"text": "@tom_doerr 135 agents, 35 curated skills, 121 plugins and 6 MCP configs all in one Claude Code toolkit is absolutely insane value \ud83e\udd2f this is the kind of community effort that makes an AI tool go from good to unstoppable. bookmarking this immediately \ud83d\udd25",
"url": "https://x.com/arpan7sarkar/status/2026746479369662551",
"author_handle": "arpan7sarkar",
"date": "2026-02-25",
"date_confidence": "high",
"engagement": null,
"relevance": 0.7,
"why_relevant": "",
"subs": {
"relevance": 70,
"recency": 100,
"engagement": 35
},
"score": 64
},
{
"id": "X7",
"text": "@tom_doerr A curated collection of Claude Skills organized by category is exactly what power users have been waiting for \ud83e\udd29 document skills, code tools, data analysis all in one repo. this is the Claude plugin store before the Claude plugin store \ud83d\udd25",
"url": "https://x.com/arpan7sarkar/status/2026745559986614335",
"author_handle": "arpan7sarkar",
"date": "2026-02-25",
"date_confidence": "high",
"engagement": null,
"relevance": 0.7,
"why_relevant": "",
"subs": {
"relevance": 70,
"recency": 100,
"engagement": 35
},
"score": 64
},
{
"id": "X2",
"text": "Claude Code\u306eSkills\u3092\u4f5c\u6210\u4f8b\u304b\u3089\u5fb9\u5e95\u7406\u89e3\u3059\u308b https://t.co/vZTseio2Ii",
"url": "https://x.com/yasuky/status/2026754773786226708",
"author_handle": "yasuky",
"date": "2026-02-25",
"date_confidence": "high",
"engagement": null,
"relevance": 0.7,
"why_relevant": "",
"subs": {
"relevance": 70,
"recency": 100,
"engagement": 35
},
"score": 64
},
{
"id": "X1",
"text": "How: I told Claude Code \"build me a CRM.\" Our builder skill generated schemas, skills, server, seed data. I pointed Claude Desktop at it. Contacts, lead scoring, pipeline reviews. All from conversation.\n\nThen \"build me a research assistant.\" Topics, sources, notes, etc. Same flow",
"url": "https://x.com/matgoldsborough/status/2026755742737510749",
"author_handle": "matgoldsborough",
"date": "2026-02-25",
"date_confidence": "high",
"engagement": {
"replies": 1
},
"relevance": 0.7,
"why_relevant": "",
"subs": {
"relevance": 70,
"recency": 100,
"engagement": 35
},
"score": 64
},
{
"id": "X10",
"text": "I can build another open claw, or claude bot, or even a messaging app in a nice prompt and although having engineering skills, I can push the code on the public repo and secure it even completely and may start marketing in a crazy way on X or maybe Reddit using a secret niche guy when writing a post.\n\nThis is the new coding for us, and shipping a nice product.",
"url": "https://x.com/0x_Kapoor/status/2026738663321907214",
"author_handle": "0x_Kapoor",
"date": "2026-02-25",
"date_confidence": "high",
"engagement": null,
"relevance": 0.7,
"why_relevant": "",
"subs": {
"relevance": 70,
"recency": 100,
"engagement": 35
},
"score": 64
},
{
"id": "X11",
"text": "1. I use https://t.co/NtFYJEiFs9 inspired by @bcherny \n2. I use https://t.co/EwnmyWNFqn to sync skills and scan skills from the marketplace\n3. I chat with claude code 4-5 times and once I am satisifed with plan, I ask it to build it.\n4. I use /code-simplifier, /memory and /parallel from my plugin above to distribute work across agents.\n5. I wait and once they're done, I use coderabbit:review to review once locally and make a PR\n\nThis process works like a gem for me.",
"url": "https://x.com/ghumare64/status/2026736176695246966",
"author_handle": "ghumare64",
"date": "2026-02-25",
"date_confidence": "high",
"engagement": {
"likes": 1
},
"relevance": 0.7,
"why_relevant": "",
"subs": {
"relevance": 70,
"recency": 100,
"engagement": 0
},
"score": 56
},
{
"id": "X12",
"text": "The 'vibe coding' revolution is forcing a reckoning.\n\nDevelopers with 15+ years experience are either embracing AI to expand their reach or watching their expertise become as valuable as COBOL skills.\n\nWhich side of history are you on?\n\nPS\nClaude Code can now write COBOL as well.",
"url": "https://x.com/hamen/status/2026734075243790396",
"author_handle": "hamen",
"date": "2026-02-25",
"date_confidence": "high",
"engagement": {
"likes": 1
},
"relevance": 0.7,
"why_relevant": "",
"subs": {
"relevance": 70,
"recency": 100,
"engagement": 0
},
"score": 56
}
],
"web": [],
"youtube": [
{
"id": "X8afcX2s2Mo",
"title": "Claude Skills: Build Your First AI Marketing Team in 16 Minutes (Claude Code)",
"url": "https://www.youtube.com/watch?v=X8afcX2s2Mo",
"channel_name": "Grace Leung",
"date": "2026-02-21",
"date_confidence": "high",
"engagement": {
"num_comments": 47,
"likes": 1900,
"views": 47474
},
"transcript_snippet": "Every marketer I know is stretch themed. Too many channels, too many deliverables, never enough time, whether you're running a team or doing it all yourself. And that's why I'm so obsessed with clock skills. You teach it how your marketing works once and it just deliver every time. And so even with just a few clock skills, you can start building a marketing team with serious leverage. And so in this video, I will show you how to build your first AI marketing team using just clock scale. So everyone is talking about cloud agents and yes they are super powerful but what enable them to be truly powerful is the expertise behind this agent and that expertise is exactly what cloud skills are. So when you're designing your skill library for your marketing team these skills contain your best marketing knowledge SOPs your proven frameworks your brand standards and agents are now the team members who use that expertise. You can equip any agents with any combination of skills from your library and that's what makes this powerful because you build this library once and cl as a team that runs on it or even you can bring it to all the AI models. So today we're building five skills that cover what most marketing team would need. Research and strategy, content creation, creatives, data analysis, and campaign presentation. Essentially, they will form a strong foundation as you build your marking skill library over time. For this demo, we'll be using Clark Cook Desktop. But if you prefer using Clark work is still fine. So, first thing is to make sure you already got the skill creator tool installed because this official skill is not building inside Clark. I will put the official skill with link and also check out this video on how to install. So here we have already set up a project folder called simple home which is the DTOC e-commerce brand and inside we have different folders and the most important is the context file. So every time when you onboard a new client or you are an in-house marketers these are the context file about the brand the product the audience etc. So you can equip cloth with all the background information and always include a clotmpd file. So these are just like a custom instructions for these project to guide how clock should navigate your folder system. So I go much deeper on how to set up these contact system inside our AI for marketing program with all the templates. So if you're interested you can find the link below to join. But for now let's create our first skill marketing research and strategy. So here inside the SOP folder we already have these SOP document that the brand is already using for marketing research and building strategy reports. So now on Clark Cook desktop, select the same project folder for this brand and then use this prompt to answer it to...",
"relevance": 0.7,
"why_relevant": "YouTube video about claude code skills and mcp servers",
"subs": {
"relevance": 70,
"recency": 86,
"engagement": 70
},
"score": 74
},
{
"id": "0J2_YGuNrDo",
"title": "Claude Code just Built me an AI Agent Team (Claude Code + Skills + MCP)",
"url": "https://www.youtube.com/watch?v=0J2_YGuNrDo",
"channel_name": "Grace Leung",
"date": "2025-12-16",
"date_confidence": "high",
"engagement": {
"num_comments": 291,
"likes": 4760,
"views": 155806
},
"transcript_snippet": "I use clot more than any other AI but even I dismiss Clark as just for developers and I was wrong after using myself I am completely so and this is too powerful to ignore if you want to use AI to build system and do work for you end to end not just chatting and copy pasting so in this video I'll share how to get started with cloud code even if you don't use for coding and most importantly how to use it to build an AI work team and system that you can actually use. Let's go. So why claw code is even needed if claw is already so powerful. So most people use cloud projects with custom instructions to turn claw into AI agents. But they are silo. You end up copy and pasting between them. You are the coordinator. But with cloud code is different. So these cloud agents share the same workspace access to the same context file. They can talk to each other and hands up work automatically each with their own token context window and claude now becomes the team lead. You give the direction they execute together and that is what we are building today. So each of these cloud code agents you create needs at least three core components all defined in one single markdown file. the role and responsibilities, the knowledge like workflow details, reusable clot agent skills, the MCP tools that they have access to. For this demo, we'll be building a five agent work and claude, our main agent will add as our team lead to coordinate between these agent. A bonus tip is when designing your AI teammate, give them specific and non-over overlapping roles to minimize the potential conflicts. Now, to get started with clark code, there are four main ways to access it. First, the local terminal. This is the raw method and will give you the best experience. Luckily, Anthroporic just launched a native installer. So, which makes it really easy to install. I'll put the official link below. And second, ID extension. So, you can add CL code to ID like VS Code or Kursa and let you run the CL code and see the project file structure side by side. Third, the cloud web app where you can click the code tab and run the CL code in the cloud, but it requires you linking to a GitHub repository. and fourth, cloud desktop app which has cloud code built-in and this is what we'll be using today. Now, whatever way you choose, I still highly recommend installing cloud code through the terminal first because the terminal gives you the most complete experience. All commands and features work fully since we're running Clark locally. So, the first thing is to set up a local project folder. So, think of it like a dedicated workspace for Clark. So, let's say I run a marketing agency and I'm setting up the TIM system. So...",
"relevance": 0.7,
"why_relevant": "YouTube video about claude code skills and mcp servers",
"subs": {
"relevance": 70,
"recency": 0,
"engagement": 100
},
"score": 61
},
{
"id": "vIUJ4Hd7be0",
"title": "Claude Code Skills - The Only Tutorial You Need",
"url": "https://www.youtube.com/watch?v=vIUJ4Hd7be0",
"channel_name": "Leon van Zyl",
"date": "2026-02-09",
"date_confidence": "high",
"engagement": {
"num_comments": 63,
"likes": 699,
"views": 25538
},
"transcript_snippet": "",
"relevance": 0.7,
"why_relevant": "YouTube video about claude code skills and mcp servers",
"subs": {
"relevance": 70,
"recency": 46,
"engagement": 54
},
"score": 59
},
{
"id": "Gqh_KdHP1Xk",
"title": "8 MCP Servers That Make Claude Code 10x Better",
"url": "https://www.youtube.com/watch?v=Gqh_KdHP1Xk",
"channel_name": "Robin Ebers",
"date": "2025-08-21",
"date_confidence": "high",
"engagement": {
"num_comments": 108,
"likes": 2719,
"views": 90281
},
"transcript_snippet": "Over 90% of all MCP servers are complete overhyped garbage. But some of them are not only worth it, they can be complete game changers. And I do not say this lightly. So after testing over 100 of them, here are the only eight MCP servers that I actually use to build apps even faster. But first, let me save you from making the same expensive mistake that I made when I started using MCP servers. Because what no one explains about them is that the more of them you install, the dumber your AI becomes. I know this sounds backwards because everyone thinks more tools equals smarter AI. But here is what's actually happening because every MCP server you add is like giving your AI yet another tool to choose from. Imagine asking someone to fix your bike, but instead of giving them a toolbox with eight essential tools, you just dump 100 tools on the floor. they will spend more time figuring out which tool to use than actually fixing your bike. That is exactly what happens with claude and chatbt too because every MCP server needs instructions in the context window or chat when you start. But new research now shows that this is actually really bad for AI. It's like trying to focus on a specific task while browsing social media. You're just pumping useless information into your brain. And that is kind of what happens too when you use too many MCP servers in your AI. So, if you want your AI to stay focused, remove all distractions. I stripped my list down to these eight MCP servers. And even then, I don't use all of them in every project. I turn them on and off because the fewer tools you give your AI, the more focused it stays. So, now let's look at my full list. Hey, what's up, guys? My name is Rob and I've been a coder for over 20 years. But now, I teach people how to build their ideas with AI in ways that even non-technical people can understand. I do this here on YouTube and inside of my AI coding blueprint, which you can check out in the description down below. Starting with number eight, almost every app that you build will need real world data. This could be stock prices, news articles, YouTube comments, or Instagram profiles. And that is where this MCP server comes in because this turns your AI into a data gathering machine. API is a marketplace for web scrapers, which you can think of as little robots that go all over the internet and grab information from any website that you want. And here is the really crazy part, because this MCP server doesn't just help you pull the data into your app. It actually makes accessing the full documentation of every web scraper available on this platform very easy. It can also help you debug them when things go wrong or check the...",
"relevance": 0.7,
"why_relevant": "YouTube video about claude code skills and mcp servers",
"subs": {
"relevance": 70,
"recency": 0,
"engagement": 84
},
"score": 56
},
{
"id": "901VMcZq8X4",
"title": "These 5 Claude Code Skills Are Your New Unfair Advantage",
"url": "https://www.youtube.com/watch?v=901VMcZq8X4",
"channel_name": "Sean Kochel",
"date": "2025-10-21",
"date_confidence": "high",
"engagement": {
"num_comments": 65,
"likes": 1356,
"views": 44602
},
"transcript_snippet": "",
"relevance": 0.7,
"why_relevant": "YouTube video about claude code skills and mcp servers",
"subs": {
"relevance": 70,
"recency": 0,
"engagement": 67
},
"score": 51
},
{
"id": "l7qVtHpctic",
"title": "Claude Code's MCP Problem Just Got Fixed",
"url": "https://www.youtube.com/watch?v=l7qVtHpctic",
"channel_name": "Kenny Liao",
"date": "2026-01-17",
"date_confidence": "high",
"engagement": {
"num_comments": 93,
"likes": 728,
"views": 21765
},
"transcript_snippet": "",
"relevance": 0.7,
"why_relevant": "YouTube video about claude code skills and mcp servers",
"subs": {
"relevance": 70,
"recency": 0,
"engagement": 54
},
"score": 47
},
{
"id": "M5CsRj6zSCA",
"title": "5 MCP Servers That Make Claude Code 10x More Powerful (Full-Stack App Build)",
"url": "https://www.youtube.com/watch?v=M5CsRj6zSCA",
"channel_name": "Eric Tech",
"date": "2025-12-12",
"date_confidence": "high",
"engagement": {
"num_comments": 30,
"likes": 266,
"views": 11286
},
"transcript_snippet": "",
"relevance": 0.7,
"why_relevant": "YouTube video about claude code skills and mcp servers",
"subs": {
"relevance": 70,
"recency": 0,
"engagement": 33
},
"score": 41
},
{
"id": "qthyl0GCpDo",
"title": "Claude Skills vs MCP: What\u2019s the Difference and When to Use Each?",
"url": "https://www.youtube.com/watch?v=qthyl0GCpDo",
"channel_name": "Postman",
"date": "2025-11-20",
"date_confidence": "high",
"engagement": {
"num_comments": 4,
"likes": 105,
"views": 7045
},
"transcript_snippet": "",
"relevance": 0.7,
"why_relevant": "YouTube video about claude code skills and mcp servers",
"subs": {
"relevance": 70,
"recency": 0,
"engagement": 12
},
"score": 35
},
{
"id": "ZroGqu7GyXM",
"title": "Claude Code Skills vs MCP vs Sub Agents: What Works for Solo Devs?",
"url": "https://www.youtube.com/watch?v=ZroGqu7GyXM",
"channel_name": "Solo Swift Crafter",
"date": "2025-10-27",
"date_confidence": "high",
"engagement": {
"num_comments": 13,
"likes": 98,
"views": 4880
},
"transcript_snippet": "",
"relevance": 0.7,
"why_relevant": "YouTube video about claude code skills and mcp servers",
"subs": {
"relevance": 70,
"recency": 0,
"engagement": 10
},
"score": 34
},
{
"id": "jzf7DQa2CAc",
"title": "How I Use Claude Code With Skills, MCP, Agents & Plugins",
"url": "https://www.youtube.com/watch?v=jzf7DQa2CAc",
"channel_name": "Matt Kuda",
"date": "2026-01-19",
"date_confidence": "high",
"engagement": {
"num_comments": 8,
"likes": 87,
"views": 2547
},
"transcript_snippet": "",
"relevance": 0.7,
"why_relevant": "YouTube video about claude code skills and mcp servers",
"subs": {
"relevance": 70,
"recency": 0,
"engagement": 0
},
"score": 31
}
],
"best_practices": [],
"prompt_pack": [],
"context_snippet_md": "# Context: Claude Code skills and MCP servers (Last 30 Days)\n\n*Generated: 2026-02-25 | Sources: both*\n\n## Key Sources\n\n- [X] @adamwathan Codex has been crushing it for me with...\n- [X] \ud83d\udea8 Anthropic just open-sourced the exact Skills lib...\n- [Reddit] We 3x'd our team's Claude Code skill usage in 2 weeks \u2014 here's how\n- [Reddit] ClaudeInOne \u2014 a full framework for Claude Code, installed in one command\n- [Reddit] Self-improvement Loop: My favorite Claude Code Skill\n- [X] Day 7 building my first B2C app\n\nBeen setting ever...\n- [Reddit] I built 12 SEO skills for Claude Code (open source)\n\n## Summary\n\n*See full report for best practices, prompt pack, and detailed sources.*\n"
}