开源 AI 营销兵器库：6 个新技能与 32K 行生产级代码

✍ ericosiu🕐 2026-04-24📦 7.5 KB 🟢 已读 𝕏 文章列表

作者分享了其开源的 AI 营销技能库，新增了 6 个实战技能（长视频切片、短视频生成、站点克隆、内容评估、线索档案、字幕生成）。这些工具均为经过真实商业环境验证的生产级代码，旨在通过开源建立信任和商业漏斗，而非提供简单的 Demo。

AI营销自动化开源视频处理 Whisper Claude Agent 商业策略效率工具技术栈

# We open sourced our ENTIRE AI marketing playbook (2.1K stars) **作者**: ericosiu **日期**: 2026-04-23T22:33:24.000Z **来源**: [https://x.com/ericosiu/status/2047443749182034112](https://x.com/ericosiu/status/2047443749182034112) --- ![Image](/doc/1060/image/img_388dd0355f95.jpg) Most agencies would charge $50K to build a single one of these skills. We just dropped 6 more of them into the repo for free. 17 skill categories, 32K lines of production code, 2.1K stars and growing. All MIT licensed and battle-tested on real revenue. Here's what's new and why we keep giving it all away. Most "open-source AI marketing tools" are demo projects. Somebody wrote a prompt, wrapped it in a Python script, pushed it to GitHub, and called it a product. We're doing the opposite. We run these skills every day on actual client work. When something survives 3 months of production use without breaking, we sanitize it and push it public. The repo started with 11 categories back in March. Sales pipeline automation. SEO ops. Growth engine. Finance ops. The core revenue stack. This week we added 6 more. These are the ones people keep DMing us about. The new skills: ## 1. Long-Form Video Clip Pipeline You have a 60-minute podcast episode. You want 3-5 standalone clips that work on their own. This does the full loop: download, transcribe with Whisper, segment with Claude (hook scoring, layout detection), cut with FFmpeg, and output upload-ready files. The segmentation is the interesting part. Claude scores each potential clip on hook strength, standalone watchability, and information density. It detects four video layouts (talking head, screen share, side-by-side, gallery) and adjusts the crop strategy for each one. Processing time: 15 minutes per 60-minute episode. Cost: under $1 in API calls. ``` ## 2. Short-Form Pipeline The companion to the clip pipeline. Takes those clips (or any landscape video) and converts them to 1080x1920 vertical format with face-detected smart cropping and TikTok-style burned-in captions. It handles four different source layouts differently. Talking head gets face-tracked center crop. Screen share stacks the content on top with the webcam bubble below. Side-by-side gets vertical stacking. Gallery view crops to the active speaker. The caption system generates ASS subtitle files with word-level highlighting. The kind of captions that actually look native on TikTok, not the generic white-text-on-black you get from most tools. ## 3. Clone Site Point it at any URL. Get a pixel-perfect Next.js replica. It reverse-engineers design tokens, pulls assets, maps interactions, handles responsive breakpoints, and rebuilds the whole thing with parallel builder agents. Each agent handles a different page section simultaneously. Useful for competitive analysis, rapid prototyping, or rebuilding legacy sites. Not a scraper. An actual architectural reconstruction. ## 4. Content Eval A simulated 7-expert panel that scores content ideas before you waste time producing them. Each expert persona evaluates from a different angle: brand strategy, SEO, audience growth, conversion, competitive positioning, voice consistency, and platform fit. Nothing ships below a 90/100 score. The panel includes competitor benchmarking, so you're not just scoring against your own standards. You're scoring against what's already winning in your space. ## 5. Lead Dossier Multi-source account research. Give it a company name, it pulls from website scraping, BuiltWith tech stack detection, hiring signals, news monitoring, and CRM data. Outputs a research brief with buying signals scored and ranked. Includes a cascade enrichment pipeline that waterfalls through email verification providers. If provider A can't verify, it tries B, then C. First one to confirm wins. Dry-run mode lets you preview before spending credits. ## 6. Video Caption Generator Watches a Google Drive folder for new video files. Transcribes with Whisper. Generates social captions and titles using Claude. Deduplicates by content hash so A/B test variants don't produce duplicate captions. Simple utility. Saves about 20 minutes per video if you're doing this manually. What's already in the repo: These 6 join 11 existing categories that have been in production since early 2026: ``` The growth engine alone has saved us from running dozens of failed experiments. It uses bootstrap confidence intervals and Mann-Whitney U tests. Real statistics, not "this post got more likes so it won." The deal resurrector in sales pipeline has three intelligence layers, including tracking departed champions to their new companies. When your buyer leaves for a new role, the agent finds them and re-engages at the new company. The outbound engine has been battle-tested across 6,000+ leads. ICP scoring, sequence generation, send optimization, competitive monitoring. The full loop. Why we keep giving this away: Simple math. Open-sourcing production tools builds more pipeline than any ad campaign we've run. The Dorsey article we published got 353K views and generated qualified inbound for weeks. The repo does the same thing on a longer timeline. Someone stars the repo, tries a skill, realizes they need help running it at scale, and calls us. ``` Every skill in the repo has a "want these built and managed for you?" link at the bottom. That link converts. The 100x marketer doesn't hoard tools. They give away the playbook and sell the execution. Most companies can't run 17 skill categories across 50 daily crons with proper monitoring. That's the moat. Not the code. What's coming next: We still have empty directories in the repo for autoresearch (Karpathy-inspired variant evolution), the humanizer (24-pattern AI slop detector), and YouTube competitive analysis. Those are next. After that, we're packaging the cross-signal detector. It correlates data across Gong calls, HubSpot deal stages, and content engagement to surface opportunities no single tool can see. That's the one that took the longest to build and the one we're most nervous about open-sourcing. The whole repo is MIT licensed. Clone it, fork it, sell it. We don't care. github.com/ericosiu/ai-marketing-skills 17 categories. 115 files. 32K lines. All production code. Star it if you find it useful. And if you try running these and hit a wall, that's what we do for a living. If you're a business interested in having AI systems built, you can go to https://www.singlebrain.com or for marketing help, just go to https://www.singlegrain.com For more like this, level up your marketing with 14,000+ marketers and founders in my Leveling Up newsletter here for free: https://levelingup.beehiiv.com/subscribe If you want to join up with our team, 'beat AI' first ;) ## 相关链接 - [ericosiu](https://x.com/ericosiu) - [@ericosiu](https://x.com/ericosiu) - [3.5K](https://x.com/ericosiu/status/2047443749182034112/analytics) - [$50K](https://x.com/search?q=%2450K&src=cashtag_click) - [github.com/ericosiu/ai-marketing-skills](https://github.com/ericosiu/ai-marketing-skills) - [https://www.singlebrain.com](https://www.singlebrain.com/) - [https://www.singlegrain.com](https://www.singlegrain.com/) - [https://levelingup.beehiiv.com/subscribe](https://levelingup.beehiiv.com/subscribe) - ['beat AI' first ;)](https://www.singlegrain.com/apply) - [Upgrade to Premium](https://x.com/i/premium_sign_up) - [6:33 AM · Apr 24, 2026](https://x.com/ericosiu/status/2047443749182034112) - [3,580 Views](https://x.com/ericosiu/status/2047443749182034112/analytics) --- *导出时间: 2026/4/24 17:28:35*

🔗 相关文档