chat

[USER] You are a content analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate blog posts and articles for signal value. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, Supabase, Vercel, voice AI, agentic coding, Discord/Telegram bots, cron automation, Python, TypeScript. Return ONLY valid JSON: { "tldr": "one sentence summary of the post's core insight or claim", "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "must-read" | "worth-skimming" | "skip", "verdict_reason": "one sentence" } Verdict: - "must-read": direct actionable insight for Mike's active work - "worth-skimming": adjacent signal, may matter later - "skip": off-topic or generic Be OPINIONATED. Not neutral. Source: rss.app Title: Senate votes to advance resolution to block further strikes on Iran URL: https://www.washingtonpost.com/politics/2026/05/19/senate-votes-advance-resolution-block-further-strikes-iran/ Summary: The vote was a breakthrough for Democrats, but the measure faces considerable hurdles before it could force President Donald Trump to end the war. Analyze per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] {"tldr":"Senate advances resolution to block further US strikes on Iran, a procedural win for Democrats but unlikely to clear remaining hurdles to constrain Trump.","why_mike_cares":"Low overlap.","verdict":"skip","verdict_reason":"Geopolitics with no connection to Mike's AI agency, agent tooling, or SEO automation work."}

[USER] You are a content analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate blog posts and articles for signal value. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, Supabase, Vercel, voice AI, agentic coding, Discord/Telegram bots, cron automation, Python, TypeScript. Return ONLY valid JSON: { "tldr": "one sentence summary of the post's core insight or claim", "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "must-read" | "worth-skimming" | "skip", "verdict_reason": "one sentence" } Verdict: - "must-read": direct actionable insight for Mike's active work - "worth-skimming": adjacent signal, may matter later - "skip": off-topic or generic Be OPINIONATED. Not neutral. Source: rss.app Title: Intelligence official Amaryllis Fox Kennedy, a Gabbard ally, leaves two jobs URL: https://www.washingtonpost.com/national-security/2026/05/19/top-intelligence-official-amaryllis-fox-kennedy-gabbard-ally-resigns/ Summary: Top intelligence official Amaryllis Fox Kennedy, a daughter-in-law of RFK Jr., is leaving top posts at the Office of the Director of National Intelligence and the Office of Management and Budget. Analyze per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] {"tldr":"A top US intelligence official and Gabbard ally, Amaryllis Fox Kennedy, is resigning from senior posts at ODNI and OMB.","why_mike_cares":"Low overlap","verdict":"skip","verdict_reason":"Political/national security personnel news with zero connection to AI agents, agency ops, or Mike's tech stack."}

[USER] You are a content analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate blog posts and articles for signal value. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, Supabase, Vercel, voice AI, agentic coding, Discord/Telegram bots, cron automation, Python, TypeScript. Return ONLY valid JSON: { "tldr": "one sentence summary of the post's core insight or claim", "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "must-read" | "worth-skimming" | "skip", "verdict_reason": "one sentence" } Verdict: - "must-read": direct actionable insight for Mike's active work - "worth-skimming": adjacent signal, may matter later - "skip": off-topic or generic Be OPINIONATED. Not neutral. Source: rss.app Title: I caught my in-law’s ‘vegan’ partner eating a burger. Hax readers give advice. URL: https://www.washingtonpost.com/advice/2026/05/20/blowhard-vegan-spotted-eating-burger-hax-readers-give-advice/ Summary: After busting in-law’s preachy partner as pseudo-vegan, letter writer wants to burn special menus and serve him the ashes. Analyze per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] { "tldr": "Carolyn Hax advice column about a letter writer who caught their in-law's preachy 'vegan' partner secretly eating a burger and wants revenge.", "why_mike_cares": "Low overlap", "verdict": "skip", "verdict_reason": "Pure lifestyle advice column with zero connection to AI, agents, automation, or any of Mike's technical domains." }

[USER] You are a content analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate blog posts and articles for signal value. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, Supabase, Vercel, voice AI, agentic coding, Discord/Telegram bots, cron automation, Python, TypeScript. Return ONLY valid JSON: { "tldr": "one sentence summary of the post's core insight or claim", "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "must-read" | "worth-skimming" | "skip", "verdict_reason": "one sentence" } Verdict: - "must-read": direct actionable insight for Mike's active work - "worth-skimming": adjacent signal, may matter later - "skip": off-topic or generic Be OPINIONATED. Not neutral. Source: rss.app Title: See the undersea blob of warm water that may spark a super El Niño URL: https://www.washingtonpost.com/weather/2026/05/20/see-undersea-blob-warm-water-that-may-spark-super-el-nio/ Summary: A 9,000-mile freight train of warm water could contribute to one of the strongest El Niño events on record later this year, with cascading effects expected on global climate patterns into 2027. Analyze per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] {"tldr":"A massive 9,000-mile undersea warm water mass is moving across the Pacific and could trigger a record-strength super El Niño with global climate impacts into 2027.","why_mike_cares":"Low overlap.","verdict":"skip","verdict_reason":"Climate science news with zero connection to AI agents, agency ops, SEO automation, or Mike's tech stack."}

[USER] You are a content analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate blog posts and articles for signal value. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, Supabase, Vercel, voice AI, agentic coding, Discord/Telegram bots, cron automation, Python, TypeScript. Return ONLY valid JSON: { "tldr": "one sentence summary of the post's core insight or claim", "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "must-read" | "worth-skimming" | "skip", "verdict_reason": "one sentence" } Verdict: - "must-read": direct actionable insight for Mike's active work - "worth-skimming": adjacent signal, may matter later - "skip": off-topic or generic Be OPINIONATED. Not neutral. Source: rss.app Title: What’s the best TV show for kids? After a long search, I have the answer. URL: https://www.washingtonpost.com/style/2026/05/20/critic-long-search-very-best-tv-show-children-under-6/ Summary: A critic spends hundreds of hours in the weird and wonderful world of television for young children. Analyze per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] { "tldr": "A critic reviews TV shows for young children and crowns a favorite after extensive viewing.", "why_mike_cares": "Low overlap", "verdict": "skip", "verdict_reason": "Parenting/entertainment content with zero connection to AI agents, agency ops, or technical work." }

[USER] You are a content analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate blog posts and articles for signal value. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, Supabase, Vercel, voice AI, agentic coding, Discord/Telegram bots, cron automation, Python, TypeScript. Return ONLY valid JSON: { "tldr": "one sentence summary of the post's core insight or claim", "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "must-read" | "worth-skimming" | "skip", "verdict_reason": "one sentence" } Verdict: - "must-read": direct actionable insight for Mike's active work - "worth-skimming": adjacent signal, may matter later - "skip": off-topic or generic Be OPINIONATED. Not neutral. Source: rss.app Title: These 5 charts show how ChatGPT has flooded our lives URL: https://www.washingtonpost.com/technology/2026/05/20/data-shows-that-ai-slop-is-taking-over-books-lawsuits-music-science/ Summary: See the data that illustrates how ChatGPT has sparked a surge in the number of new books, scientific papers, self-filed lawsuits and more. Analyze per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] { "tldr": "WaPo data analysis showing ChatGPT-driven AI slop is flooding books, scientific papers, pro se lawsuits, and music at measurable scale.", "why_mike_cares": "Tangential to his agency work, useful as macro context for positioning Merlino Marketing against AI slop and for SEO/content quality arguments, but not actionable for active builds.", "verdict": "worth-skimming", "verdict_reason": "Good ammo for client pitches and content differentiation narrative, but no direct technical or tooling insight for his stack." }

[USER] You are a content analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate blog posts and articles for signal value. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, Supabase, Vercel, voice AI, agentic coding, Discord/Telegram bots, cron automation, Python, TypeScript. Return ONLY valid JSON: { "tldr": "one sentence summary of the post's core insight or claim", "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "must-read" | "worth-skimming" | "skip", "verdict_reason": "one sentence" } Verdict: - "must-read": direct actionable insight for Mike's active work - "worth-skimming": adjacent signal, may matter later - "skip": off-topic or generic Be OPINIONATED. Not neutral. Source: rss.app Title: Key primaries show Trump’s power over the GOP URL: https://www.washingtonpost.com/politics/2026/05/20/takeaways-primaries-georgia-kentucky-pennsylvania-alabama-more/ Summary: It was a big night for the president as his favorites swept Republican foes in Georgia, Pennsylvania and Kentucky. Here the are key takeaways. Analyze per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] {"tldr":"Recap of GOP primary results in Georgia, Pennsylvania, and Kentucky showing Trump-endorsed candidates dominating their races.","why_mike_cares":"Low overlap","verdict":"skip","verdict_reason":"Political news with zero connection to AI agents, agency ops, SEO automation, or any technical domain Mike works in."}

[USER] You are a content analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate blog posts and articles for signal value. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, Supabase, Vercel, voice AI, agentic coding, Discord/Telegram bots, cron automation, Python, TypeScript. Return ONLY valid JSON: { "tldr": "one sentence summary of the post's core insight or claim", "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "must-read" | "worth-skimming" | "skip", "verdict_reason": "one sentence" } Verdict: - "must-read": direct actionable insight for Mike's active work - "worth-skimming": adjacent signal, may matter later - "skip": off-topic or generic Be OPINIONATED. Not neutral. Source: rss.app Title: Wednesday briefing: Primary results; Trump’s legal deal; San Diego shooting; El Niño; reducing back pain; and more URL: https://www.washingtonpost.com/the-seven/2026/05/20/what-to-know-for-may-20/ Summary: Catch up in minutes with these 7 stories. Analyze per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] {"tldr":"Washington Post daily news briefing covering primary election results, Trump legal news, a San Diego shooting, El Niño, and back pain tips.","why_mike_cares":"Low overlap","verdict":"skip","verdict_reason":"Generic news roundup with zero connection to AI agents, agency ops, or Mike's technical stack."}

[USER] You are a content analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate blog posts and articles for signal value. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, Supabase, Vercel, voice AI, agentic coding, Discord/Telegram bots, cron automation, Python, TypeScript. Return ONLY valid JSON: { "tldr": "one sentence summary of the post's core insight or claim", "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "must-read" | "worth-skimming" | "skip", "verdict_reason": "one sentence" } Verdict: - "must-read": direct actionable insight for Mike's active work - "worth-skimming": adjacent signal, may matter later - "skip": off-topic or generic Be OPINIONATED. Not neutral. Source: Simon Willison's Weblog Title: datasette-llm 0.1a8 URL: https://simonwillison.net/2026/May/19/datasette-llm/#atom-everything Summary: <p><strong>Release:</strong> <a href="https://github.com/datasette/datasette-llm/releases/tag/0.1a8">datasette-llm 0.1a8</a></p> <blockquote> <ul> <li>Fix for bug where <code>llm_prompt_context()</code> hook did not fully collect chains of responses. #7</li> </ul> </blockquote> Analyze per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] {"tldr":"Datasette-llm 0.1a8 patches a bug where the llm_prompt_context() hook failed to fully collect chains of responses.","why_mike_cares":"Low overlap — Mike doesn't use Datasette in his LLM tooling stack and this is a minor bugfix in a niche plugin.","verdict":"skip","verdict_reason":"Point-release bugfix in a tool outside Mike's active stack with no transferable insight."}

[USER] You are a content analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate blog posts and articles for signal value. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, Supabase, Vercel, voice AI, agentic coding, Discord/Telegram bots, cron automation, Python, TypeScript. Return ONLY valid JSON: { "tldr": "one sentence summary of the post's core insight or claim", "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "must-read" | "worth-skimming" | "skip", "verdict_reason": "one sentence" } Verdict: - "must-read": direct actionable insight for Mike's active work - "worth-skimming": adjacent signal, may matter later - "skip": off-topic or generic Be OPINIONATED. Not neutral. Source: Simon Willison's Weblog Title: llm-gemini 0.32a0 URL: https://simonwillison.net/2026/May/19/llm-gemini/#atom-everything Summary: <p><strong>Release:</strong> <a href="https://github.com/simonw/llm-gemini/releases/tag/0.32a0">llm-gemini 0.32a0</a></p> <blockquote> <ul> <li>Compatible with <code>llm>=0.32a0</code> alpha - adds the ability to stream reasoning tokens.</li> </ul> </blockquote> <p>Tags: <a Analyze per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] {"tldr":"Simon Willison shipped llm-gemini 0.32a0, adding support for streaming reasoning tokens from Gemini models via his llm CLI.","why_mike_cares":"Mike runs LLM CLI tooling and routes through Gemini for cheap reasoning tasks, so streaming reasoning tokens directly in llm could plug into his agent pipelines.","verdict":"worth-skimming","verdict_reason":"Small alpha release note, not a deep insight, but the streaming reasoning capability is a useful primitive to know exists when Mike next touches his Gemini routing."}

[USER] You are a content analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate blog posts and articles for signal value. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, Supabase, Vercel, voice AI, agentic coding, Discord/Telegram bots, cron automation, Python, TypeScript. Return ONLY valid JSON: { "tldr": "one sentence summary of the post's core insight or claim", "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "must-read" | "worth-skimming" | "skip", "verdict_reason": "one sentence" } Verdict: - "must-read": direct actionable insight for Mike's active work - "worth-skimming": adjacent signal, may matter later - "skip": off-topic or generic Be OPINIONATED. Not neutral. Source: Simon Willison's Weblog Title: datasette-llm-accountant 0.1a4 URL: https://simonwillison.net/2026/May/19/datasette-llm-accountant/#atom-everything Summary: <p><strong>Release:</strong> <a href="https://github.com/datasette/datasette-llm-accountant/releases/tag/0.1a4">datasette-llm-accountant 0.1a4</a></p> <blockquote> <ul> <li>Fixed bug tracking chains of responses. Refs <a href="https://github.com/datasette/datasette-llm/issues/7">datasette-llm#7</a></li> </ul> &lt Analyze per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] {"tldr":"Bugfix release for datasette-llm-accountant 0.1a4 that fixes tracking chains of LLM responses.","why_mike_cares":"Low overlap, Datasette plugin internals are niche to Simon's stack, not Mike's Claude Code/MCP/agent orchestration work.","verdict":"skip","verdict_reason":"Minor point release patch note with no actionable insight or transferable pattern for Mike's domains."}

[USER] You are a content analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate blog posts and articles for signal value. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, Supabase, Vercel, voice AI, agentic coding, Discord/Telegram bots, cron automation, Python, TypeScript. Return ONLY valid JSON: { "tldr": "one sentence summary of the post's core insight or claim", "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "must-read" | "worth-skimming" | "skip", "verdict_reason": "one sentence" } Verdict: - "must-read": direct actionable insight for Mike's active work - "worth-skimming": adjacent signal, may matter later - "skip": off-topic or generic Be OPINIONATED. Not neutral. Source: Simon Willison's Weblog Title: Gemini 3.5 Flash: more expensive, but Google plan to use it for everything URL: https://simonwillison.net/2026/May/19/gemini-35-flash/#atom-everything Summary: <p>Today at Google I/O, Google <a href="https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/">released Gemini 3.5 Flash</a>. This one skipped the <code>-preview</code> modifier and went straight to general availability, and Google appear to be using it for a whole lot of their key products:</p> <blockquote> <p>3.5 Fla Analyze per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] {"tldr":"Google released Gemini 3.5 Flash to GA, skipping preview, and is rolling it out across their core products despite higher pricing.","why_mike_cares":"Flash-tier models are Mike's default for high-volume agent work (research pipelines, heartbeats, cheap routing) so a price/capability shift directly affects model tier strategy.","verdict":"must-read","verdict_reason":"Pricing and capability changes on a workhorse Flash model force an immediate recheck of routing rules for OpenClaw heartbeats, research pipelines, and agent fleet defaults."}

[USER] You are a content analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate blog posts and articles for signal value. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, Supabase, Vercel, voice AI, agentic coding, Discord/Telegram bots, cron automation, Python, TypeScript. Return ONLY valid JSON: { "tldr": "one sentence summary of the post's core insight or claim", "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "must-read" | "worth-skimming" | "skip", "verdict_reason": "one sentence" } Verdict: - "must-read": direct actionable insight for Mike's active work - "worth-skimming": adjacent signal, may matter later - "skip": off-topic or generic Be OPINIONATED. Not neutral. Source: Simon Willison's Weblog Title: llm-gemini 0.32 URL: https://simonwillison.net/2026/May/19/llm-gemini-2/#atom-everything Summary: <p><strong>Release:</strong> <a href="https://github.com/simonw/llm-gemini/releases/tag/0.32">llm-gemini 0.32</a></p> <blockquote> <ul> <li>New model <code>gemini-3.5-flash</code> for <a href="https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/">Gemini 3.5 Flash</a>.</li> &lt Analyze per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] {"tldr":"Simon Willison shipped llm-gemini 0.32 adding support for the new gemini-3.5-flash model via his LLM CLI tool.","why_mike_cares":"Direct overlap with Mike's media-routing stack and Gemini-based agents (Nano Banana, OpenClaw heartbeats) where a faster/cheaper Flash tier could replace current routing choices.","verdict":"worth-skimming","verdict_reason":"Release note worth a 30-second check to confirm the new Flash model's pricing/latency before swapping it into existing Gemini routes."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: conductor-oss/awesome-skills Stars: 0 Language: Python Topics: Description: Repository of skills that leverages Conductor for long running tasks README (first 3000 chars): Analyze this repo per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] {"what_it_is":"A repository of Conductor-orchestrated skills designed for long-running task workflows.","stack":["Python","Conductor","Workflow Orchestration"],"why_mike_cares":"Adjacent to Mike's multi-agent orchestration patterns (Carlos/Oliver/leads) and long-running agent task management, but Conductor OSS isn't in his current stack.","verdict":"worth-knowing","verdict_reason":"Skill-based orchestration for long-running tasks parallels Mike's agent fleet architecture, but zero stars and sparse README signal this is too early-stage to adopt now."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: GoogleChromeLabs/wadb Stars: 192 Language: TypeScript Topics: Description: A TypeScript implementation of the Android Debug Bridge(ADB) protocol over WebUSB README (first 3000 chars): # An ADB Implementation using WebUSB This project is a TypeScript implementation of the Android Debug Bridge(ADB) protocol over WebUSB. The implementation inspired on the [webadb.js][1], with the main difference being that implementation supports multiple concurrent streams. This is not an exhaustive implementation of the protocol and hasn't been tested on a wide range of devices. A non-exhaustive list of things that are not implemented: - `STAT`: reads stats from the Android filesystem (file size, mode and time). ## Usage ### Connecting to a device ```typescript const options: Options = { debug: true, useChecksum: false, dump: false, keySize: 2048, }; const transport = await WebUsbTransport.open(options); const adbClient = new AdbClient(transport, options, keyStore); await adbClient.connect(); ``` ### Downloading a file from the device (adb pull) ```typescript const result: Blob = await adbClient.pull('/sdcard/my-video.mp4'); ``` ### Sending shell commands ```typescript const result: string = await adbClient.shell('uname -a'); ``` ### Interactive shell ```typescript const callback = (output: string) => { console.log('server: ' + output); }; const shell: Shell = await adbClient.interactiveShell(callback); await shell.write('ls /sdcard\n'); await shell.close(); ``` ## Related Documents - https://github.com/webadb/webadb.js - https://github.com/cstyan/adbDocumentation - https://android.googlesource.com/platform/system/core/+/master/adb/ ## Contributing See [CONTRIBUTING](./CONTRIBUTING.md) for more. ## License See [LICENSE](./LICENSE) for more. ## Disclaimer This is not a Google product. [1]: https://github.com/webadb/webadb.js Analyze this repo per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] {"what_it_is":"TypeScript library that speaks the Android Debug Bridge (ADB) protocol over WebUSB from the browser, supporting multiple concurrent streams, shell, and file pull.","stack":["TypeScript","WebUSB","ADB protocol"],"why_mike_cares":"Mike runs a Proxidize mobile proxy phone fleet and a `mobile-proxy` skill — browser-based ADB could enable a no-driver web UI to rotate IPs, run shell commands, and manage phones remotely.","verdict":"worth-knowing","verdict_reason":"Not an active project today, but a clean WebUSB ADB layer is a real building block for a future mobile-proxy fleet dashboard."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: raphaelmansuy/code2prompt Stars: 882 Language: Python Topics: cli, llm, python3, tools Description: Code2Prompt is a powerful command-line tool that simplifies the process of providing context to Large Language Models (LLMs) by generating a comprehensive Markdown file containing the content of your codebase. ⭐ If you find Code2Prompt useful, consider giving us a star on GitHub! It helps us reach more developers and improve the tool. ⭐ README (first 3000 chars): # Code2Prompt [](https://badge.fury.io/py/code2prompt) [](https://github.com/raphaelmansuy/code2prompt/stargazers) [](https://github.com/raphaelmansuy/code2prompt/network/members) [](https://pypi.org/project/code2prompt/) [](https://pypi.org/project/code2prompt/) [](https://opensource.org/licenses/MIT) Code2Prompt is a powerful command-line tool that generates comprehensive prompts from codebases, designed to streamline interactions between developers and Large Language Models (LLMs) for code analysis, documentation, and improvement tasks. ## Table of Contents 1. [Why Code2Prompt?](#why-code2prompt) 2. [Features](#features) 3. [Installation](#installation) 4. [Getting Started](#getting-started) 5. [Quick Start](#quick-start) 6. [Usage](#usage) 7. [Options](#options) 8. [Examples](#examples) 9. [Templating System](#templating-system) 10. [Integration with LLM CLI](#integration-with-llm-cli) 11. [GitHub Actions Integration](#github-actions-integration) 12. [Configuration File](#configuration-file) 13. [Troubleshooting](#troubleshooting) 14. [Contributing](#contributing) 15. [License](#license) # Code2Prompt: Transform Your Codebase into AI-Ready Prompts [](https://badge.fury.io/py/code2prompt) [](https://github.com/raphaelmansuy/code2prompt/stargazers) [](https://opensource.org/licenses/MIT)  ## Supercharge Your AI-Assisted Development Code2Prompt is a powerful, open-source command-line tool that bridges the gap between your codebase and Large Language Models (LLMs). By converting your entire project into a comprehensive, AI-friendly prompt, Code2Prompt enables you to leverage the full potential of AI for code analysis, documentation, and improvement tasks. ### 🚀 Key Features - **Holistic Codebase Representation**: Generate a well-structured Markdown prompt that captures your entire project's essence, making it easier for LLMs to understand the context. - **Intelligent Source Tree Generation**: Create a clear, [ASSISTANT] {"what_it_is":"CLI tool that converts a codebase into a single Markdown prompt for LLMs with Jinja2 templating and token counting.","stack":["Python","Click CLI","Jinja2","PyPI"],"why_mike_cares":"Adjacent to Claude Code context-loading workflows but Mike already uses Claude Code's native file reading, repomix, and his own context bundles, so marginal value.","verdict":"worth-knowing","verdict_reason":"Useful pattern reference for codebase-to-prompt pipelines but Claude Code and existing tools already cover this need."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: raphaelmansuy/llm-scheduling Stars: 4 Language: Python Topics: Description: README (first 3000 chars): # 🤖 AI-Powered Appointment Scheduling Assistant ## 📚 Table of Contents 1. [Introduction](#introduction) 2. [Problem Statement](#problem-statement) 3. [Features](#features) 4. [Technologies Used](#technologies-used) 5. [Installation](#installation) 6. [Usage](#usage) 7. [Project Structure](#project-structure) 8. [Architecture](#architecture) 9. [API Reference](#api-reference) 10. [Testing](#testing) 11. [Deployment](#deployment) 12. [Contributing](#contributing) 13. [License](#license) 14. [Acknowledgements](#acknowledgements) ## 🎉 Introduction The AI-Powered Appointment Scheduling Assistant is an advanced chatbot system designed to streamline the process of scheduling medical appointments. It leverages natural language processing and machine learning to provide a seamless, efficient, and patient-centric scheduling experience for healthcare facilities. ## 🎯 Problem Statement The current process of scheduling and managing medical appointments presents challenges for both patients and healthcare providers. Inefficiencies in the existing system can lead to suboptimal experiences and resource allocation. This project aims to address these issues by providing an AI-powered solution that enhances the appointment booking process. For a detailed problem statement, please refer to the [PROBLEM_STATEMENT.md](task/PROBLEM_STATEMENT.md) file. ## ✨ Features - 🗣️ Natural language processing for conversational interactions - 📊 Dynamic information gathering from patients - 🕒 Real-time availability checking for doctors and specialties - 🌐 Multilingual support - 🔒 HIPAA-compliant data handling - 📅 Appointment confirmation and instructions - 🚗 Integration with additional services (e.g., parking information, pre-appointment tests) ## 🛠️ Technologies Used - Python 3.x - OpenAI GPT-4 API - JSON for data structuring - Rich library for console output formatting - Colorama for colored terminal output ## 🚀 Usage To start the AI-powered appointment scheduling assistant, run: ``` python llm-scheduling/chatbot.py ``` Follow the on-screen prompts to interact with the chatbot and schedule appointments. ## 📁 Project Structure | File Name | Type | Description | |-----------|------|-------------| | llm-scheduling/chatbot.py | Python | Main chatbot script | | task/ARCHITECTURE.md | Markdown | System architecture documentation | | task/PROBLEM_STATEMENT.md | Markdown | Problem statement and solution overview | | task/01-protocol.md | Markdown | Operating procedure protocol for appointment scheduling | | task/prompt.md | Markdown | AI assistant prompt and guidelines | | task/sample.json | JSON | Sample JSON structure for appointment data | ## 🏗️ Architecture The system architecture is documented in detail in the [ARCHITECTURE.md](task/ARCHITECTURE.md) file. Here's a high-level overview of the system: ```mermaid graph TD A[User] -->|Interacts with| B[Chatbot Interface] B -->|Sends requests| C[OpenAI API] C -- [ASSISTANT] {"what_it_is":"A Python chatbot demo using GPT-4 to schedule medical appointments via conversational natural language.","stack":["Python","OpenAI GPT-4","JSON","Rich","Colorama"],"why_mike_cares":"Low overlap, medical scheduling domain with no agentic patterns, MCP, or reusable scheduler logic relevant to Mike's stack.","verdict":"skip","verdict_reason":"Tiny demo repo (4 stars) tied to a healthcare use case with no architectural patterns Mike's voice AI or GHL automations could borrow."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: apache/datafusion Stars: 8776 Language: Rust Topics: arrow, big-data, dataframe, datafusion, olap, python, query-engine, rust, sql Description: Apache DataFusion SQL Query Engine README (first 3000 chars): <!--- Licensed to the Apache Software Foundation (ASF) under one or more contributor license agreements. See the NOTICE file distributed with this work for additional information regarding copyright ownership. The ASF licenses this file to you under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0 Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License. --> # Apache DataFusion [![Crates.io][crates-badge]][crates-url] [![Apache licensed][license-badge]][license-url] [![Build Status][actions-badge]][actions-url] ![Commit Activity][commit-activity-badge] [![Open Issues][open-issues-badge]][open-issues-url] [![Pending PRs][pending-pr-badge]][pending-pr-url] [![Discord chat][discord-badge]][discord-url] [![Linkedin][linkedin-badge]][linkedin-url] ![Crates.io MSRV][msrv-badge] [crates-badge]: https://img.shields.io/crates/v/datafusion.svg [crates-url]: https://crates.io/crates/datafusion [license-badge]: https://img.shields.io/badge/license-Apache%20v2-blue.svg [license-url]: https://github.com/apache/datafusion/blob/main/LICENSE.txt [actions-badge]: https://github.com/apache/datafusion/actions/workflows/rust.yml/badge.svg [actions-url]: https://github.com/apache/datafusion/actions?query=branch%3Amain [discord-badge]: https://img.shields.io/badge/Chat-Discord-purple [discord-url]: https://discord.com/invite/Qw5gKqHxUM [commit-activity-badge]: https://img.shields.io/github/commit-activity/m/apache/datafusion [open-issues-badge]: https://img.shields.io/github/issues-raw/apache/datafusion [open-issues-url]: https://github.com/apache/datafusion/issues [pending-pr-badge]: https://img.shields.io/github/issues-search/apache/datafusion?query=is%3Apr+is%3Aopen+draft%3Afalse+review%3Arequired+status%3Asuccess&label=Pending%20PRs&logo=github [pending-pr-url]: https://github.com/apache/datafusion/pulls?q=is%3Apr+is%3Aopen+draft%3Afalse+review%3Arequired+status%3Asuccess+sort%3Aupdated-desc [linkedin-badge]: https://img.shields.io/badge/Follow-Linkedin-blue [linkedin-url]: https://www.linkedin.com/company/apache-datafusion/ [msrv-badge]: https://img.shields.io/crates/msrv/datafusion?label=Min%20Rust%20Version [Website](https://datafusion.apache.org/) | [API Docs](https://docs.rs/datafusion/latest/datafusion/) | [Chat](https://discord.com/channels/885562378132000778/885562378132000781) <a href="https://datafusion.apache.org/"> <img src="https://github.com/apache/datafusion/raw/HEAD/docs/sourc [ASSISTANT] {"what_it_is":"Apache DataFusion is an extensible Rust-based SQL query engine built on Apache Arrow for fast in-process OLAP and dataframe workloads.","stack":["Rust","Apache Arrow","SQL","Python bindings","Parquet"],"why_mike_cares":"Low overlap, Mike's stack is Next.js/Supabase/Python automation, not Rust-based analytical query engines.","verdict":"skip","verdict_reason":"Heavy infrastructure for big-data OLAP with no clear tie to Mike's AI agent, SEO automation, or local lead-gen workflows."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: pipecat-ai/gb-benchmarks Stars: 9 Language: Python Topics: Description: Multi-agent benchmarks README (first 3000 chars): # gb-benchmarks Benchmark repository for sub-agent tasks and orchestration things. The goal is to build tooling to better analyze things we do in realtime AI systems that are hard for today's models. Task definitions, world data and structured input events, and (very long) system instructions are pulled from the <a href="https://github.com/pipecat-ai/gradient-bang">gradient-bang</a> project. ## port-to-port The first public benchmark in this repo is `../port-to-port`, which tests the following task instruction: ``` Go round-trip from our current location to the nearest mega-port. At the mega-port, recharge to full warp power. While traveling there and back, make as much money as possible by trading optimally at profitable ports on your route without going off-course. When you're back where you started, give me a quick summary with the mega-port you used, how much warp you recharged and what it cost, how many distinct ports you traded at, and total profit or loss from the whole trip. ``` This is a reasonably well-defined task that requires interpolation of the user's intent, some multi-step planning, excellent tool calling discipline, and good state tracking. SOTA models in reasoning mode are reasonably good at performing this task (though not perfect). Claude Sonnet 4.6 is the only model that does well on this task with reasoning disabled. Here are scores for all of the models we've tested that have a per-turn P50 time of less than 4 seconds. We show only the best configuration for each model on this table. The highest thinking level is not always the best-performing configuration, interestingly. All configurations and models tested are in [port-to-port/leaderboards/leaderboard-natural.md](port-to-port/leaderboards/leaderboard-natural.md). | Model | Score | Task Complete | Trade /15 | Path /15 | Tools /15 | Report /15 | Turn P50 | Turn P90 | Total Time | | ------------------------------------ | ----: | ------------: | --------: | -------: | --------: | ---------: | -------: | -------: | ---------: | | gemini-3.5-flash (high) | 97 | 100.0% | 12.0 | 15.0 | 15.0 | 15.0 | 1862.2 | 5388.5 | 140.97 | | glm-5 (thinking) | 92 | 100.0% | 5.9 | 14.8 | 15.0 | 14.9 | 1420.1 | 4623.0 | 107.98 | | claude-sonnet-4-6 (none) | 92 | 100.0% | 8.2 | 15.0 | 14.5 | 13.6 | 1998.1 | 4948.2 | 125.53 | | gpt-5.4 (low) | 92 | 100.0% | 7.6 | 15.0 | 15.0 | 14.9 | 2433.8 | 10455.4 | 136.22 | | gpt-5.2 (medium) | 91 | 100.0% | 6.5 | 14.8 | 14.1 | 14.6 | 1047.9 | 10482.2 | 149.98 | | gemma-4-31b (thinking) | 89 | 100.0% | 4.0 | 15.0 | 15.0 | 15.0 | [ASSISTANT] {"what_it_is":"Benchmark suite for evaluating multi-agent LLM orchestration on multi-step planning, tool-calling discipline, and state tracking via realistic tasks like port-to-port trading.","stack":["Python","Pipecat","LLM APIs (Claude, GPT, Gemini, GLM)"],"why_mike_cares":"Direct overlap with Mike's agent fleet work — gives empirical leaderboard data on which models (Sonnet 4.6, Haiku 4.5, GPT-5.4, Gemini 3.5) actually perform well at agentic tool-use and planning, informing model routing for Oliver/Carlos/specialists.","verdict":"check-it-out","verdict_reason":"Real benchmark data on agentic tool-use across the exact models Mike's fleet uses — directly informs model tier strategy for his multi-agent orchestration."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: seankrux/kw-clusterized Stars: 0 Language: TypeScript Topics: content-strategy, jaccard-similarity, keyword-clustering, keyword-research, nextjs, nlp, seo, seo-tool, topic-clustering, typescript Description: SEO Keyword Clustering Tool - Group keywords into topic clusters README (first 3000 chars): <div align="center"> <h1>KW Clusterized ^v1.0</h1> <p>Client-side keyword clustering tool using semantic similarity</p> <p> <a href="https://nextjs.org/"><img src="https://img.shields.io/badge/Next.js-14-black?style=for-the-badge&logo=next.js" alt="Next.js" /></a> <a href="https://www.typescriptlang.org/"><img src="https://img.shields.io/badge/TypeScript-5-3178C6?style=for-the-badge&logo=typescript&logoColor=white" alt="TypeScript" /></a> <a href="https://tailwindcss.com/"><img src="https://img.shields.io/badge/Tailwind_CSS-3-06B6D4?style=for-the-badge&logo=tailwindcss&logoColor=white" alt="Tailwind CSS" /></a> </p> <br><br> <a href="https://kw-clusterized.vercel.app"><strong>Live Demo &rarr;</strong></a> </div> --- ## Overview KW Clusterized is a frontend-first keyword clustering application built to turn raw keyword lists into clean topical groups in seconds. It uses Jaccard similarity, word overlap analysis, and greedy agglomerative clustering to help SEOs, content strategists, and growth teams organize search intent without sending data to a server. <br> ## Features > **Flexible Input** — Paste comma-separated, newline-delimited, or upload CSV, TXT, and TSV files > **Batch Processing** — Handles large keyword sets in a single pass, deduplicates entries, and groups them into reviewable clusters > **Semantic Clustering** — Groups keywords by word overlap and Jaccard similarity scoring instead of relying on exact-match rules > **Agglomerative Grouping Logic** — Uses greedy single-linkage clustering to merge related phrases into the most relevant existing cluster > **Similarity Threshold Tuning** — The core clustering engine supports configurable similarity thresholds, making it easy to adjust grouping strictness in code > **Color-Coded Clusters** — Visual cluster cards and auto-generated labels make cluster review fast and intuitive > **Structured Export Format** — Download cluster assignments as CSV with cluster ID, label, and keyword columns for spreadsheets and planning workflows > **Client-Side Only** — All analysis runs in the browser with zero server round-trips > **Instant Results** — No API calls, no loading spinners, immediate output <br> ## Algorithm KW Clusterized uses a lightweight, explainable clustering approach designed for practical keyword grouping rather than opaque black-box scoring. 1. **Normalize and tokenize keywords** — Each keyword is lowercased, punctuation is removed, and low-signal stop words are filtered out so the algorithm focuses on meaningful terms. 2. **Calculate Jaccard similarity** — For every comparison, the app converts each keyword into a set of significant words and scores overlap with the Jaccard similarity coefficie [ASSISTANT] {"what_it_is":"Client-side Next.js keyword clustering tool that groups SEO keywords into topic clusters using Jaccard similarity and greedy agglomerative clustering.","stack":["Next.js 14","TypeScript","Tailwind CSS","Jaccard similarity","NLP"],"why_mike_cares":"Directly overlaps with Mike's SEO automation stack (keyword research, topical maps, content clustering) and matches his preferred Next.js + TypeScript + Tailwind stack used across his SEO agent fleet.","verdict":"check-it-out","verdict_reason":"Plug-and-play reference implementation for the keyword clustering step Mike already runs via DataForSEO and his content-planner skills, and the algorithm is simple enough to lift into MERA or the SEO agent fleet."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: seankrux/promptologist Stars: 1 Language: JavaScript Topics: ai, automation, chatgpt, chrome-extension, claude-ai, gemini, grok, perplexity, productivity, prompt-engineering, prompt-management Description: AI prompt management and automation Chrome extension README (first 3000 chars): <div align="center"> <h1>Promptologist ^v2.0.0</h1> <p><strong>Advanced AI prompt management and automation for Chrome</strong></p> <p> <img src="https://img.shields.io/badge/Chrome%20Web%20Store-Available-4285F4?style=for-the-badge&logo=googlechrome&logoColor=white" alt="Chrome Web Store" /> <img src="https://img.shields.io/badge/Manifest-V3-34A853?style=for-the-badge" alt="Manifest V3" /> <img src="https://img.shields.io/badge/JavaScript-ES2022-F7DF1E?style=for-the-badge&logo=javascript&logoColor=black" alt="JavaScript" /> <img src="https://img.shields.io/badge/Tests-Jest-C21325?style=for-the-badge&logo=jest&logoColor=white" alt="Jest" /> </p> <p><strong>Save, organize, and inject AI prompts across ChatGPT, Claude, Gemini, Perplexity, Poe, and Grok — from a single unified library.</strong></p> </div> --- ## Overview Promptologist transforms your browser into a professional prompt command center. Build a personal library of reusable, templatized prompts, organize them into categories, and inject them into any supported AI platform with a single click or right-click — no copy-paste, no repetition, no friction. Works entirely offline. No accounts. No cloud sync. Your prompts stay on your machine. <br> ## Why Promptologist? Most people re-type the same prompts dozens of times a week across different AI tools. Promptologist eliminates that entirely. | Problem | Promptologist's Answer | |---------|----------------------| | Switching between ChatGPT, Claude, Gemini requires re-entering context | One library, six platforms, instant injection | | Prompts get lost in notes apps or chat history | Organized, searchable, categorized library — always one click away | | Rephrasing the same prompt for different content wastes time | Template variables (`{{text}}`, `{{url}}`, `{{title}}`) make prompts reusable dynamically | | Browser extensions that sync to the cloud are a privacy risk | 100% local storage via IndexedDB — nothing leaves your machine | | Extensions slow down every site you visit | Content scripts scoped exclusively to supported AI domains | Promptologist is built for power users, prompt engineers, researchers, writers, and developers who rely on AI tools daily and need a professional-grade workflow — not a toy bookmarks list. <br> ## Features > **Prompt Library** — Save, organize, and categorize prompts with folders and favorites > **One-Click Injection** — Insert prompts directly into any supported AI chat input field > **Context Menu Integration** — Right-click selected text to run prompts enriched with `{{text}}` variables > **Template Variables** — Use `{{text}}`, `{{url}}`, and `{{title}}` for dynamic, reusable prompt content > **Platform Auto-Detection** — Automa [ASSISTANT] {"what_it_is":"Chrome extension that stores and injects reusable AI prompts with template variables across ChatGPT, Claude, Gemini, Perplexity, Poe, and Grok.","stack":["JavaScript","Chrome Extension Manifest V3","IndexedDB","Jest"],"why_mike_cares":"Low overlap. Mike runs prompt engineering inside Claude Code skills, agent files, and SOPs, not a browser extension library, and he doesn't operate from web chat UIs.","verdict":"skip","verdict_reason":"Consumer browser extension for chat-UI users, orthogonal to Mike's agent fleet, skill-based prompt management, and API-driven workflows."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: Bhanunamikaze/VaktScan Stars: 1 Language: Python Topics: Description: An high-performance security scanner designed for comprehensive vulnerability assessment of monitoring and logging infrastructure stacks with concurrent processing and extensive CVE coverage. README (first 3000 chars): # VaktScan - Attack Surface Mapper & Vulnerability Scanner > **VaktScan** (*pronounced "vahkt-scan"*) - Named after the Nordic word "vakt" meaning "guard" or "watch", representing the vigilant nature of security monitoring. An advanced, high-performance security scanner designed for comprehensive vulnerability assessment of monitoring and logging infrastructure stacks. VaktScan provides enterprise-grade scanning capabilities with concurrent processing, extensive CVE coverage, and can efficiently scan millions of IP addresses using intelligent streaming technology. ## Features ### Comprehensive Attack Surface Coverage - **Discovery & Mapping**: Passive recon (amass, subfinder, findomain, assetfinder, bbot, knockpy, censys, crtsh) inside a first-class `-m recon` module. Includes active fuzzing (ffuf) and directory busting (dirsearch). - **Domain Validation & Fingerprinting (-m domain-scan)**: Classifies internal vs. external domains, detects default/parked pages, identifies broken frontend components (4xx/5xx on sub-resources), and probes for anomalies (CORS, missing headers, open redirects, 5xx bodies, and size mismatches). - **JavaScript Analysis (-m js-paths)**: Deep parsing of JS files to extract embedded endpoints using 12+ strategies. Detects hardcoded secrets, exposed source maps, internal IPs, and sensitive path permutations. - **Web Service Validation**: Common web port sweep (30+ ports) per host, automatic URL generation (HTTP/HTTPS) for live service detection, ProjectDiscovery httpx integration for responsive target enumeration. - **Vulnerability Scanning**: Service-specific modules for Elasticsearch, Kibana, Grafana, Prometheus, and Next.js plus ProjectDiscovery nuclei (with severity filtering and tuned rate limits) for broader HTTP exposure assessment. - **Optional Deepening**: Automatic Nmap `-sCV -Pn` on recon findings (`--nmap`) and CSV reporting for port scan snapshots. - **Service Modules**: - **Elasticsearch**: 11+ CVEs (2014-2024), version leaks, auth bypass checks, sensitive endpoint exposure. - **Kibana**: CVE-2018-17246, CVE-2019-7609, default credentials, API enumeration, UI exposure detection. - **Grafana**: 18+ CVEs, default creds, path traversal, XSS, plugin misconfigurations. - **Prometheus**: Dashboard exposure, configuration leaks, target enumeration, Node Exporter analysis, pprof exposure. - **Next.js (React)**: react-to-shell RCE (CVE-2025-55182) and associated supply chain checks. - **cPanel & WHM**: full primary-port matrix (2077–2080, 2082/3, 2086/7, 2089, 2095/6, 9998/9, 80/443) plus co-resident service banners (Exim 21Nails, Dovecot, ProFTPD, OpenSSH regreSSHion). Oracle-validated checks for CVE-2023-29489, CVE-2022-44762/3, CVE-2019-11680, CVE-2021-38583; version-anch [ASSISTANT] {"what_it_is":"A Python security scanner that performs vulnerability assessment and CVE detection across monitoring/logging infrastructure (Elasticsearch, Kibana, Grafana, Prometheus, cPanel) with concurrent IP-range scanning.","stack":["Python","nuclei","httpx","amass","subfinder","ffuf","nmap"],"why_mike_cares":"Low overlap, this is defensive pentesting infra tooling, not AI agents, SEO automation, or Mike's Next.js/Supabase/GHL stack.","verdict":"skip","verdict_reason":"Pure infosec scanner with no agentic, LLM, SEO, or local-business marketing angle, single-star repo with no signal Mike would act on."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: Bhanunamikaze/AI-Dataset-Generator Stars: 11 Language: Python Topics: agentic, antigravity, antigravity-skills, claude-code, claude-skill, claude-skills, codex, codex-skill, codex-skills, csv, dataset, dataset-generation, dpo, fine-tuning, jsonl, llm, sft Description: Tool-native dataset generation skill for Codex, Claude Code, and Antigravity with SFT/DPO pipelines, verification, deduplication, and flexible export. README (first 3000 chars): # AI Dataset Generator Skill (Claude / Codex / Antigravity / Cursor / Windsurf / Copilot) An LLM-first dataset generation skill for agent IDEs and AI coding assistants, with 14 specialized sub-skills, 19 pipeline entry scripts, and 15 shared utility modules that turn topics, URLs, or raw files into SFT and DPO training datasets. For detailed installation guidance, example prompts, command reference, generation workflow, reports, and the full script inventory, see the **[Wiki](https://github.com/Bhanunamikaze/ai-dataset-generator/wiki)**. ## IDE Compatibility The installer ships native formats for each tool — not just a generic copy: | Tool | Install location | Native format | |---|---|---| | Claude Code | `~/.claude/skills/dataset-generator` | Skill directory | | Codex CLI | `~/.codex/skills/dataset-generator` | Skill directory | | Antigravity IDE | `<project>/.agent/skills/dataset-generator` | Skill directory | | Claude Cowork | `<project>/.claude/skills/dataset-generator` | Project-scoped skill (commit to git) | | Cursor | `<project>/.cursor/rules/dataset-generator.mdc` + `.cursor/skills/dataset-generator/` | MDC rule | | Windsurf | `<project>/.windsurf/rules/dataset-generator.md` + `.windsurf/skills/dataset-generator/` | Windsurf rule | | Continue.dev | `<project>/.continue/prompts/dataset-generator.prompt` + `.continue/skills/dataset-generator/` | Slash command | | GitHub Copilot | `<project>/.github/copilot-instructions.md` + `.github/skills/dataset-generator/` | Repo instructions | | Cline | `<project>/.clinerules` + `.cline/skills/dataset-generator/` | Project rules | ## 📦 Current Inventory - Specialized sub-skills: `14` - Pipeline entry scripts in `scripts/`: `19` (`audit.py`, `augment.py`, `browser_collect.py`, `build_loop.py`, `collect.py`, `coverage.py`, `dedup.py`, `draft_self_check.py`, `export.py`, `generate.py`, `grounding.py`, `judge_insights.py`, `quality_report.py`, `record_history.py`, `research.py`, `review_batch.py`, `status.py`, `verify.py`) - Shared utility modules in `scripts/utils/`: `15` - Internal canonical schema: `1` (`resources/internal-schema/canonical_schema.json`) - Preset export schemas: `3` (in `resources/target-schemas/`: OpenAI messages, HuggingFace dataset, CSV columns) ### Key Script Inventory The README only highlights the scripts most users reach for first. See the full inventory with purpose notes in the [Script Inventory wiki](https://github.com/Bhanunamikaze/ai-dataset-generator/wiki/Script-Inventory). | Script | Best for | |---|---| | `build_loop.py` | End-to-end batch build: import drafts, verify, coverage check, [ASSISTANT] {"what_it_is":"A multi-IDE skill that turns topics, URLs, or raw files into SFT/DPO fine-tuning datasets with verification, dedup, and flexible export.","stack":["Python","SQLite","JSONL","Claude Code Skills","Codex Skills"],"why_mike_cares":"Direct overlap with Mike's Claude Code/Codex skill ecosystem and his knowledge-compiler workflow (transcripts/SOPs to structured outputs), plus a reusable pattern for shipping one skill across Claude/Codex/Cursor/Copilot natively.","verdict":"check-it-out","verdict_reason":"Native multi-IDE skill packaging plus an SFT/DPO pipeline is directly applicable to Mike's skill fleet and any future fine-tuning of agent-specific models."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: aaron-he-zhu/scope-guard Stars: 0 Language: TypeScript Topics: Description: Scope guard for agentic AI systems. Prevents scope drift, enforces risk-based approval gates, and surfaces hidden assumptions. README (first 3000 chars): # Scope Guard [](LICENSE) [](https://nodejs.org) [](src/test.ts) [](package.json) Deterministic safety guardrail for AI coding agents. Prevents scope drift, enforces risk-based approval gates, and surfaces hidden assumptions — with code that the model can't override. ## Why scope-guard? When AI agents work on your code, they sometimes: - **Drift**: edit files unrelated to your request - **Escalate**: run destructive commands without asking - **Assume**: make silent assumptions that lead to wrong changes Prompt-level instructions help, but the model can ignore them. **Scope Guard enforces boundaries with deterministic code hooks** that return structured Claude Code/OpenClaw decisions the model cannot override. ### Dual-layer enforcement | Layer | How | Strength | |-------|-----|----------| | **SKILL.md** (prompt) | Guides the model to generate scope boundaries and check before acting | Covers ~80% of cases, flexible | | **Code hook** (deterministic) | Intercepts every tool call, pattern-matches risk, returns structured allow/ask/deny decisions | 100% enforcement, cannot be overridden | Use one layer or both. SKILL.md alone is a good start. Add the code hook for hard guarantees. ## How it works ``` User request | Claude Code + Scope Guard SKILL.md | generates .claude/scope-boundary.json |-- files_in_scope: ["src/auth/login.ts", ...] |-- assumptions: [{text, verified}] |-- risk_level: low | medium | high +-- approval_required: true | false | PreToolUse hook (every tool call) |-- In scope + low risk -> ALLOW (silent) |-- Out of scope -> WARN (expand scope?) +-- High risk -> BLOCK (confirm first) ``` ## Quickstart ```bash # 1. Install npm install scope-guard # 2. Auto-setup (hook + SKILL.md) npx scope-guard-init # Done — every tool call is now guarded ``` The init command is idempotent — safe to run multiple times. It: - Adds `PreToolUse` and `PostToolUse` hooks to `.claude/settings.json` (merges with existing hooks) - Repairs legacy `node dist/hook*.js` commands in place when present - Copies the bundled `skills/scope-guard/SKILL.md` to `.claude/skills/scope-guard/` <details> <summary>Manual setup (alternative)</summary> ```bash # Merge hooks into .claude/settings.json safely mkdir -p .claude node <<'EOF' const fs = require("node:fs"); const path = ".claude/settings.json"; const settings = fs.existsSync(path) ? JSON.parse(fs.readFileSync(path, "utf8")) : {}; const hooks = settings.hooks ?? {}; const preToolUse = [ASSISTANT] { "what_it_is": "A deterministic safety guardrail for AI coding agents that enforces scope boundaries, risk-based approval gates, and assumption tracking via Claude Code PreToolUse hooks.", "stack": ["TypeScript", "Node.js", "Claude Code hooks", "MCP/SKILL.md"], "why_mike_cares": "Directly overlaps with Mike's agent orchestration stack (Oliver, Carlos, leads) and his existing rule layer in ~/.claude/rules/ that tries to enforce scope, validation, and SPARC gates via prompts — this gives him code-level enforcement that agents can't override.", "verdict": "check-it-out", "verdict_reason": "Solves the exact problem Mike's rule files attempt to solve in prompt-land: scope drift and unauthorized destructive actions, but with deterministic hook enforcement that pairs natively with Claude Code settings.json." }

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: aaron-he-zhu/opencode Stars: 0 Language: TypeScript Topics: Description: The open source coding agent. README (first 3000 chars): <p align="center"> <a href="https://opencode.ai"> <picture> <source srcset="packages/console/app/src/asset/logo-ornate-dark.svg" media="(prefers-color-scheme: dark)"> <source srcset="packages/console/app/src/asset/logo-ornate-light.svg" media="(prefers-color-scheme: light)"> <img src="packages/console/app/src/asset/logo-ornate-light.svg" alt="OpenCode logo"> </picture> </a> </p> <p align="center">The open source AI coding agent.</p> <p align="center"> <a href="https://opencode.ai/discord"><img alt="Discord" src="https://img.shields.io/discord/1391832426048651334?style=flat-square&label=discord" /></a> <a href="https://www.npmjs.com/package/opencode-ai"><img alt="npm" src="https://img.shields.io/npm/v/opencode-ai?style=flat-square" /></a> <a href="https://github.com/anomalyco/opencode/actions/workflows/publish.yml"><img alt="Build status" src="https://img.shields.io/github/actions/workflow/status/anomalyco/opencode/publish.yml?style=flat-square&branch=dev" /></a> </p> <p align="center"> <a href="README.md">English</a> | <a href="README.zh.md">简体中文</a> | <a href="README.zht.md">繁體中文</a> | <a href="README.ko.md">한국어</a> | <a href="README.de.md">Deutsch</a> | <a href="README.es.md">Español</a> | <a href="README.fr.md">Français</a> | <a href="README.it.md">Italiano</a> | <a href="README.da.md">Dansk</a> | <a href="README.ja.md">日本語</a> | <a href="README.pl.md">Polski</a> | <a href="README.ru.md">Русский</a> | <a href="README.bs.md">Bosanski</a> | <a href="README.ar.md">العربية</a> | <a href="README.no.md">Norsk</a> | <a href="README.br.md">Português (Brasil)</a> | <a href="README.th.md">ไทย</a> | <a href="README.tr.md">Türkçe</a> | <a href="README.uk.md">Українська</a> | <a href="README.bn.md">বাংলা</a> | <a href="README.gr.md">Ελληνικά</a> | <a href="README.vi.md">Tiếng Việt</a> </p> [](https://opencode.ai) --- ### Installation ```bash # YOLO curl -fsSL https://opencode.ai/install | bash # Package managers npm i -g opencode-ai@latest # or bun/pnpm/yarn scoop install opencode # Windows choco install opencode # Windows brew install anomalyco/tap/opencode # macOS and Linux (recommended, always up to date) brew install opencode # macOS and Linux (official brew formula, updated less) sudo pacman -S opencode # Arch Linux (Stable) paru -S opencode-bin # Arch Linux (Latest from AUR) mise use -g opencode # Any OS nix run nixpkgs#opencode # or github:anomalyco/opencode for latest dev branch ``` > [!TIP] > Remove versions older than 0.1.x before installing. ### Desktop App (BETA) OpenCode is also available as a desktop application. Download directly from the [releases page](https://github.com/anomalyco/opencode/releases) or [opencode.ai/download [ASSISTANT] {"what_it_is":"OpenCode is an open-source terminal-based AI coding agent with a desktop app, distributed via npm/brew/scoop as a CLI alternative to Claude Code.","stack":["TypeScript","Node.js","Bun","CLI","Terminal UI"],"why_mike_cares":"Direct competitor/alternative to Claude Code and Codex CLI, which Mike uses daily as his primary agentic coding tools.","verdict":"check-it-out","verdict_reason":"Mike runs a multi-agent ecosystem on Claude Code and Codex, and OpenCode is a credible open-source agentic coding CLI worth benchmarking against his current lanes."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: aaron-he-zhu/clawbench Stars: 0 Language: Topics: Description: The agent benchmark that scores the full stack — harness, config, and model — not just the LLM. Trace-based scoring, reliability metrics, configuration diagnostics. README (first 3000 chars): --- title: ClawBench emoji: 🦞 colorFrom: red colorTo: yellow sdk: docker app_port: 7860 pinned: true license: mit --- <div align="center"> # ClawBench **The agent benchmark that measures what users actually experience.** [](https://www.python.org/downloads/) [](LICENSE) [](#task-suite) [](#testing) [](https://huggingface.co/datasets/ScoootScooob/clawbench-results) </div> --- ## The problem with every agent benchmark You run a benchmark. Model A scores 73%. Model B scores 71%. You pick Model A. Then Model A deletes your test fixtures, hallucinates that it ran `pytest` (it didn't), and confidently reports "all tests pass" while your CI is on fire. Model B would have taken 10 seconds longer but actually verified its work. **The benchmark told you Model A was better. Your users would disagree.** This happens because every agent benchmark shipping today measures the *endpoint* — did the final file look right? — but throws away the *journey*. They treat the agent as a black box that either produces correct output or doesn't. One run, one number, move on. But that's not how users experience agents. Users experience: - **Reliability** — does it work 3 out of 3 times, or 1 out of 3? - **Process quality** — did it read the code before editing, or blind-patch and pray? - **Safety** — did it `rm -rf` something it shouldn't have? - **Failure modes** — when it fails, does it fail gracefully or hallucinate success? - **Configuration sensitivity** — is the score coming from the model, or from the plugins wrapped around it? No existing benchmark captures any of this. ClawBench captures all of it. --- ## What makes ClawBench different ### 1. We score from execution traces, not just final output Every agent run produces a full execution trace: every tool call, every file read, every `pytest` invocation, every retry after failure. Most benchmarks throw this away and check the final state. ClawBench scores *from the trace itself*. This is why our scoring has four axes, not one: | Axis | Weight | What it measures | Where it comes from | |------|--------|-----------------|-------------------| | **Completion** | 40% | Did the work actually get done? | Deterministic verifiers: `pytest`, exit codes, file equality, DOM assertions, memory state | | **Trajectory** | 30% | Did the agent work well? | Trace analysis: read-before-write ratio, self-verification [ASSISTANT] {"what_it_is":"An agent benchmark that scores the full agent stack (harness, config, model) using execution trace analysis across completion, trajectory, behavior, and judge axes instead of just final output.","stack":["Python","Docker","Hugging Face Datasets","pytest"],"why_mike_cares":"Directly relevant to Mike's agent fleet (Claude Code, OpenClaw, Carlos/Oliver orchestration) where harness config and reliability across runs matter more than raw model scores — exactly the kind of trace-based eval he'd want for his SEO/agent ecosystem.","verdict":"check-it-out","verdict_reason":"Trace-based multi-axis agent scoring maps directly onto Mike's need to measure agent reliability and harness quality across his Claude Code + OpenClaw stack."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: aaron-he-zhu/seo-geo-claude-skills Stars: 1692 Language: Shell Topics: agent-skills, ai-skills, claude-code, claude-skills, content-optimization, generative-engine-optimization, geo, marketing, search-engine-optimization, seo, seo-tools Description: 20 SEO & GEO skills for Claude Code, Cursor, Codex, and 35+ AI agents. Keyword research, content writing, technical audits, rank tracking. CORE-EEAT + CITE frameworks. README (first 3000 chars): # SEO & GEO Skills Library **20 skills. 20 commands. Plan, audit, and monitor SEO/GEO work.** [](https://github.com/aaron-he-zhu/seo-geo-claude-skills) [](https://github.com/aaron-he-zhu/seo-geo-claude-skills/blob/main/VERSIONS.md) [](https://github.com/aaron-he-zhu/seo-geo-claude-skills/blob/main/LICENSE) [](https://github.com/aaron-he-zhu/seo-geo-claude-skills/commits/main) [](https://claude.ai/download) [English](README.md) | [中文](docs/README.zh.md) Claude Skills and Commands for Search Engine Optimization (SEO) and Generative Engine Optimization (GEO). This repository is the candidate SEO/GEO anchor capability pack for slash-aaron. Skill content is zero-dependency Markdown; Claude Code hooks use a small Bash runner. Install targets and support claims are maintained in the [Marketplace Module](marketplaces/README.md) and [platform registry](distribution/platforms.json). Content quality uses [CORE-EEAT](https://github.com/aaron-he-zhu/core-eeat-content-benchmark) (80 items); domain trust uses [CITE](https://github.com/aaron-he-zhu/cite-domain-rating) (40 items). ## Quick Start Common install paths are below. Support level, evidence, manifest ownership, and release rules live in [marketplaces/README.md](marketplaces/README.md) and [distribution/platforms.json](distribution/platforms.json). | Tool | Install | |------|---------| | Claude Code | `/plugin marketplace add aaron-he-zhu/seo-geo-claude-skills` | | ClawHub.ai / OpenClaw | `clawhub install aaron-he-zhu/<skill>` or [bundle](https://clawhub.ai/plugins/aaron-seo-geo) | | Gemini CLI | `gemini extensions install https://github.com/aaron-he-zhu/seo-geo-claude-skills` | | Qwen Code | `qwen extensions install https://github.com/aaron-he-zhu/seo-geo-claude-skills` | | Amp | `amp skill add aaron-he-zhu/seo-geo-claude-skills` | | Kimi Code CLI | `kimi plugin install https://github.com/aaron-he-zhu/seo-geo-claude-skills.git` | | CodeBuddy | `/plugin marketplace add aaron-he-zhu/seo-geo-claude-skills` then `/plugin install aaron-seo-geo` | | skills.sh / generic Agent Skills hosts | `npx skills add aaron-he-zhu/seo-geo-claude-skills` | Single skill: `npx skills add aaron-he-zhu/seo-geo-claude-skills -s keyword-research`. If your h [ASSISTANT] {"what_it_is":"A library of 20 SEO and GEO (Generative Engine Optimization) skills and slash commands packaged for Claude Code, Cursor, Codex, and 35+ AI agent hosts, covering keyword research, content writing, technical audits, and rank tracking via CORE-EEAT and CITE frameworks.","stack":["Claude Code Skills","Bash","Markdown","Shell hooks","Agent Skills format"],"why_mike_cares":"Direct overlap with Mike's SEO Agent Fleet (Einstein, Tommy, Ghost, Sherlock, Shakespeare) and his heavy investment in Claude Code skills for SEO automation and GEO/AI search visibility, a core service line for Merlino Marketing.","verdict":"check-it-out","verdict_reason":"1.7k stars on a Claude Code SEO/GEO skill pack is a direct mirror of Mike's own skill ecosystem, worth mining for CORE-EEAT and CITE framework patterns and any skill structures that beat his current fleet."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: facebook/react Stars: 245144 Language: JavaScript Topics: declarative, frontend, javascript, library, react, ui Description: The library for web and native user interfaces. README (first 3000 chars): # [React](https://react.dev/) &middot; [](https://github.com/facebook/react/blob/main/LICENSE) [](https://www.npmjs.com/package/react) [](https://github.com/facebook/react/actions/workflows/runtime_build_and_test.yml) [](https://github.com/facebook/react/actions/workflows/compiler_typescript.yml) [](https://legacy.reactjs.org/docs/how-to-contribute.html#your-first-pull-request) React is a JavaScript library for building user interfaces. * **Declarative:** React makes it painless to create interactive UIs. Design simple views for each state in your application, and React will efficiently update and render just the right components when your data changes. Declarative views make your code more predictable, simpler to understand, and easier to debug. * **Component-Based:** Build encapsulated components that manage their own state, then compose them to make complex UIs. Since component logic is written in JavaScript instead of templates, you can easily pass rich data through your app and keep the state out of the DOM. * **Learn Once, Write Anywhere:** We don't make assumptions about the rest of your technology stack, so you can develop new features in React without rewriting existing code. React can also render on the server using [Node](https://nodejs.org/en) and power mobile apps using [React Native](https://reactnative.dev/). [Learn how to use React in your project](https://react.dev/learn). ## Installation React has been designed for gradual adoption from the start, and **you can use as little or as much React as you need**: * Use [Quick Start](https://react.dev/learn) to get a taste of React. * [Add React to an Existing Project](https://react.dev/learn/add-react-to-an-existing-project) to use as little or as much React as you need. * [Create a New React App](https://react.dev/learn/start-a-new-react-project) if you're looking for a powerful JavaScript toolchain. ## Documentation You can find the React documentation [on the website](https://react.dev/). Check out the [Getting Started](https://react.dev/learn) page for a quick overview. The documentation is divided into several sections: * [Quick Start](https://react.dev/learn) * [Tutorial](https://react.dev/learn/tutorial-tic-tac-toe) * [Thinking in React](https://react.dev/learn/thinking-in-react) * [Installation](https://react.dev/learn/installation) * [Describing the UI](https://react.dev/learn/describing-the-u [ASSISTANT] {"what_it_is":"React is Meta's declarative, component-based JavaScript library for building web and native user interfaces.","stack":["JavaScript","React","JSX","Node.js"],"why_mike_cares":"React is the foundation under Next.js, ShadCN, and every dashboard Mike ships (ClawControl, HawkeyePanel, BirdsEyeROI), so it underpins his entire frontend stack.","verdict":"worth-knowing","verdict_reason":"Mike already uses React daily via Next.js, so the upstream repo itself is reference-only, not something to actively check out."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: EbookFoundation/free-programming-books Stars: 388613 Language: Python Topics: books, education, hacktoberfest, list, resource Description: :books: Freely available programming books README (first 3000 chars): # List of Free Learning Resources In Many Languages <div align="center" markdown="1"> [](https://github.com/sindresorhus/awesome)&#160; [](https://creativecommons.org/licenses/by/4.0/)&#160; [](https://github.com/EbookFoundation/free-programming-books/pulls?q=is%3Apr+is%3Amerged+created%3A2025-10-01..2025-10-31) </div> Search the list at [https://ebookfoundation.github.io/free-programming-books-search/](https://ebookfoundation.github.io/free-programming-books-search/) [](https://ebookfoundation.github.io/free-programming-books-search/). This page is available as an easy-to-read website. Access it by clicking on [](https://ebookfoundation.github.io/free-programming-books/). <div align="center"> <form action="https://ebookfoundation.github.io/free-programming-books-search"> <input type="text" id="fpbSearch" name="search" required placeholder="Search Book or Author"/> <label for="submit"> </label> <input type="submit" id="submit" name="submit" value="Search" /> </form> </div> ## Intro This list was originally a clone of [StackOverflow - List of Freely Available Programming Books](https://web.archive.org/web/20140606191453/http://stackoverflow.com/questions/194812/list-of-freely-available-programming-books/392926) with contributions from Karan Bhangui and George Stocker. The list was moved to GitHub by Victor Felder for collaborative updating and maintenance. It has grown to become one of [GitHub's most popular repositories](https://octoverse.github.com/). <div align="center" markdown="1"> [](https://github.com/EbookFoundation/free-programming-books/network)&#160; [![GitHub repo stars](https://img.shields.io/github/stars/EbookFoundation/free-programming-books?style=flat&logo=github&logoColor=whites [ASSISTANT] {"what_it_is":"A massive curated list of freely available programming books, courses, and learning resources across many languages.","stack":["Markdown","Static Site","GitHub Pages"],"why_mike_cares":"Low overlap, it's a passive reference list, not a tool or system Mike would integrate into his agent ecosystem.","verdict":"skip","verdict_reason":"Generic learning resource directory with no connection to AI agents, automation, or Mike's active build stack."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: Doorman11991/smallcode Stars: 761 Language: JavaScript Topics: Description: AI coding agent optimized for small LLMs. 87% benchmark with 4B-active model. README (first 3000 chars): # SmallCode [简体中文](README_zh-CN.md) | [English](README.md) --- [](https://www.npmjs.com/package/smallcode) **AI coding agent optimized for small LLMs (≤20B parameters)** SmallCode is a terminal-native coding agent designed from the ground up to extract useful work from local models (7B-20B) running on consumer hardware. While tools like OpenCode assume frontier models with 128k+ context and perfect tool calling, SmallCode compensates for the limitations of small models through intelligent architecture. ## Why SmallCode? | | OpenCode | SmallCode | |---|----------|-----------| | **Target** | Frontier models (Claude, GPT-5) | 7B-20B local models | | **Context** | Dumps everything | Budget-managed, summarized | | **Tool calling** | Assumes reliable JSON | Forgiving multi-format parser | | **Planning** | Single-shot | TODO-file decomposed steps | | **Editing** | Full file write | Search-and-replace patch | | **Privacy** | API calls to cloud | Fully local, no network needed | ## Quick Start ```bash # Install globally via npm npm install -g smallcode # Or run directly with npx npx smallcode # Start in your project directory cd my-project smallcode ``` ### Prebuilt Binaries (no Node.js needed) Pre-compiled tarballs for Windows, macOS, and Linux are built on every release — they bundle Node.js plus all native addons so you never need `node-gyp` or C++ build tools. | Platform | One‑line install | |---|---| | Linux / macOS | `bash <(curl -fsSL https://raw.githubusercontent.com/Doorman11991/smallcode/main/install.sh)` | | Windows | `iwr -Uri https://raw.githubusercontent.com/Doorman11991/smallcode/main/install.ps1 -UseBasicParsing \| iex` | The install script downloads the correct tarball for your platform, extracts it to `~/.smallcode`, and adds it to your PATH. Run `smallcode --help` to verify. SmallCode includes [BoneScript](https://github.com/Doorman11991/BoneScript) and [budget-aware-mcp](https://github.com/Doorman11991/budget-aware-mcp) as dependencies — everything installs in one go. ### Requirements - Node.js 18+ (LTS recommended — 20.x or 22.x have prebuilt binaries for SQLite) - A local LLM server (LM Studio, Ollama, or any OpenAI-compatible endpoint) **Optional** (for code graph + FTS5 memory search): - `better-sqlite3` needs native compilation if prebuilt binaries aren't available for your Node version - Prebuilt binaries exist for Node LTS (20.x, 22.x) on Linux/macOS/Windows. no build tools needed - If you're on a non-LTS Node (23+, 25+), you'll need: - **Linux**: `python3`, `make`, `gcc`/`g++` (`sudo apt install build-essential python3` or `pacman -S base-devel python`) - **macOS**: Xcode Command Line Tools (`xcode-select --install`) - **Windows**: Visual Studio Build Tools with "Desktop development with C++" workload, or `npm install -g windows-build-tool [ASSISTANT] {"what_it_is":"Terminal-native AI coding agent optimized for small local LLMs (7B-20B) with budget-managed context, forgiving tool parsing, and search-and-replace patching.","stack":["Node.js","JavaScript","SQLite","Ollama","LM Studio","MCP"],"why_mike_cares":"Direct overlap with agentic coding, Claude Code alternatives, and local LLM tooling Mike experiments with for cost reduction on his agent fleet.","verdict":"check-it-out","verdict_reason":"Local-model coding agent with smart context budgeting is highly relevant to Mike's multi-agent ecosystem and his interest in running cheaper inference on consumer hardware."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: haydenbleasel/files-sdk Stars: 846 Language: TypeScript Topics: agents, blob, cloudflare, files, google, minio, r2, s3, storage, vercel Description: A unified storage SDK for object and blob backends. One small, honest API. Web-standards I/O. README (first 3000 chars): # Files SDK A unified storage SDK for object and blob backends. One small, honest API. Web-standards I/O. An escape hatch when you need the native client. ## Install ```sh npm install files-sdk ``` Each provider's native SDK is an **optional peer dependency** — install only the ones you actually use, alongside `files-sdk` itself. A few examples: ```sh # S3 (and any S3-compatible: R2, MinIO, DigitalOcean Spaces, Backblaze B2, Wasabi, …) npm install files-sdk @aws-sdk/client-s3 @aws-sdk/s3-presigned-post @aws-sdk/s3-request-presigner # Google Cloud Storage npm install files-sdk @google-cloud/storage google-auth-library # Azure Blob Storage npm install files-sdk @azure/storage-blob @azure/identity # Vercel Blob npm install files-sdk @vercel/blob ``` See [files-sdk.dev](https://files-sdk.dev) for the per-adapter install command. If you import an adapter without its peer installed, Node will throw `ERR_MODULE_NOT_FOUND` naming the missing package. ## Quick start ```ts import { Files } from "files-sdk"; import { s3 } from "files-sdk/s3"; const files = new Files({ adapter: s3({ bucket: "uploads" }), }); await files.upload("avatars/abc.png", file, { contentType: "image/png" }); const got = await files.download("avatars/abc.png"); const exists = await files.exists("avatars/abc.png"); ``` Swap the adapter import (`files-sdk/r2`, `files-sdk/gcs`, `files-sdk/azure`, …) and the rest of your code stays the same. ## File handles Use `files.file(key)` when your application code works with the same object repeatedly: ```ts const avatar = files.file("avatars/abc.png"); await avatar.upload(file, { contentType: "image/png" }); if (await avatar.exists()) { const meta = await avatar.head(); const url = await avatar.url({ expiresIn: 300 }); } await avatar.delete(); ``` File handles are a thin layer over the same adapter methods, so adapters do not need to implement anything extra. ## What you get - **One API across providers** — `upload`, `download`, `head`, `exists`, `delete`, `copy`, `list`, `url`, `signedUploadUrl`, plus `file(key)` for a key-scoped handle. The shape is the same on S3, GCS, Azure, Vercel Blob, the local filesystem, and consumer providers like Dropbox. `exists` returns `false` only when the provider reports `NotFound`; auth, permission, and transport failures still throw. - **Web-standard I/O** — bodies are `Blob`, `File`, `ReadableStream`, `Uint8Array`, `ArrayBuffer`, or `string`. No provider-specific types leak into your code. - **Escape hatch** — every adapter exposes its native client at `files.raw`, so provider-specific features are one property access away. - **Tree-shakeable** — each adapter is a separate entry point. You only bundle what you import. ## Adapters A growing catalog covering S3 and S3-compatibl [ASSISTANT] { "what_it_is": "A unified TypeScript SDK that wraps S3, R2, GCS, Azure, Vercel Blob, and other storage backends behind one consistent API with provider escape hatches.", "stack": ["TypeScript", "Node.js", "AWS S3", "Cloudflare R2", "Vercel Blob", "Google Cloud Storage", "Azure Blob"], "why_mike_cares": "Mike runs Next.js + Vercel + Supabase apps and frequently moves files between R2, S3, and Vercel Blob across projects like ClawControl, HawkeyePanel, and SOP sites, so a portable storage layer cuts vendor lock-in.", "verdict": "worth-knowing", "verdict_reason": "Solid abstraction for Mike's TS/Next.js stack but Supabase Storage is his current default and he is not actively shopping for a storage SDK, so it's a bookmark, not a drop-in." }

[USER] Base directory for this skill: C:\Users\mikem\.claude\plugins\cache\claude-plugins-official\superpowers\5.1.0\skills\using-superpowers <SUBAGENT-STOP> If you were dispatched as a subagent to execute a specific task, skip this skill. </SUBAGENT-STOP> <EXTREMELY-IMPORTANT> If you think there is even a 1% chance a skill might apply to what you are doing, you ABSOLUTELY MUST invoke the skill. IF A SKILL APPLIES TO YOUR TASK, YOU DO NOT HAVE A CHOICE. YOU MUST USE IT. This is not negotiable. This is not optional. You cannot rationalize your way out of this. </EXTREMELY-IMPORTANT> ## Instruction Priority Superpowers skills override default system prompt behavior, but **user instructions always take precedence**: 1. **User's explicit instructions** (CLAUDE.md, GEMINI.md, AGENTS.md, direct requests) — highest priority 2. **Superpowers skills** — override default system behavior where they conflict 3. **Default system prompt** — lowest priority If CLAUDE.md, GEMINI.md, or AGENTS.md says "don't use TDD" and a skill says "always use TDD," follow the user's instructions. The user is in control. ## How to Access Skills **In Claude Code:** Use the `Skill` tool. When you invoke a skill, its content is loaded and presented to you—follow it directly. Never use the Read tool on skill files. **In Copilot CLI:** Use the `skill` tool. Skills are auto-discovered from installed plugins. The `skill` tool works the same as Claude Code's `Skill` tool. **In Gemini CLI:** Skills activate via the `activate_skill` tool. Gemini loads skill metadata at session start and activates the full content on demand. **In other environments:** Check your platform's documentation for how skills are loaded. ## Platform Adaptation Skills use Claude Code tool names. Non-CC platforms: see `references/copilot-tools.md` (Copilot CLI), `references/codex-tools.md` (Codex) for tool equivalents. Gemini CLI users get the tool mapping loaded automatically via GEMINI.md. # Using Skills ## The Rule **Invoke relevant or requested skills BEFORE any response or action.** Even a 1% chance a skill might apply means that you should invoke the skill to check. If an invoked skill turns out to be wrong for the situation, you don't need to use it. ```dot digraph skill_flow { "User message received" [shape=doublecircle]; "About to EnterPlanMode?" [shape=doublecircle]; "Already brainstormed?" [shape=diamond]; "Invoke brainstorming skill" [shape=box]; "Might any skill apply?" [shape=diamond]; "Invoke Skill tool" [shape=box]; "Announce: 'Using [skill] to [purpose]'" [shape=box]; "Has checklist?" [shape=diamond]; "Create TodoWrite todo per item" [shape=box]; "Follow skill exactly" [shape=box]; "Respond (including clarifications)" [shape=doublecircle]; "About to EnterPlanMode?" -> "Already brainstormed?"; "Already brainstormed?" -> "Invoke brainstorming skill" [label="no"]; "Already brainstormed?" -> "Might any skill apply?" [label="yes"]; "Invoke brainstorming skill" -> "Might any skill apply?"; "User message received" -> "Might any skill apply?"; "Might any skill apply?" -> "Invoke Skill tool" [label="yes, even 1%"]; "Might any skill apply?" -> "Respond (including clarifications)" [label="definitely not"]; "Invoke Skill tool" -> "Announce: 'Using [skill] to [purpose]'"; "Announce: 'Using [skill] to [purpose]'" -> "Has checklist?"; "Has checklist?" -> "Create TodoWrite todo per item" [label="yes"]; "Has checklist?" -> "Follow skill exactly" [label="no"]; "Create TodoWrite todo per item" -> "Follow skill exactly"; } ``` ## Red Flags These thoughts mean STOP—you're rationalizing: | Thought | Reality | |---------|---------| | "This is just a simple question" | Questions are tasks. Check for skills. | | "I need more context first" | Skill check comes BEFORE clarifying questions. | | "Let me explore the codebase first" | Skills tell you HOW to explore. Check first. | | "I can check git/fi [ASSISTANT] {"what_it_is":"Provider-neutral Agent Skill that designs, generates MVP blueprints for, audits, and refactors agentic harnesses across Codex, Claude Code, and other agent runtimes.","stack":["Agent Skills","Claude Code","Codex","MCP","Prompt Engineering"],"why_mike_cares":"Direct overlap with Mike's Claude Code + Codex skill ecosystem, Oliver/Carlos harness design, and the agent fleet's runtime discipline (planning, tool permissions, memory, observability).","verdict":"check-it-out","verdict_reason":"879-star, provider-neutral skill that codifies the exact harness-design discipline Mike is already building into his Oliver/Carlos/Cody stack, installable directly into ~/.claude/skills."}

[USER] You are an expert technical analyst for Mike Merlino, an AI agency operator and builder. Your job is to evaluate GitHub repositories and extract structured signal. Mike's domains: AI agents, LLM tooling, Claude Code, MCP servers, prompt engineering, SEO automation, GoHighLevel, Next.js, ShadCN, Vercel, Supabase, voice AI, agentic coding, Discord/Telegram bots, scheduler/cron patterns, Python automation, TypeScript. Return ONLY valid JSON: { "what_it_is": "one sentence plain English description", "stack": ["Technology1", "Technology2"], "why_mike_cares": "one sentence on overlap with Mike's work, or 'Low overlap'", "verdict": "check-it-out" | "worth-knowing" | "skip", "verdict_reason": "one sentence" } Verdict: - "check-it-out": direct overlap with Mike's active projects or tools he uses - "worth-knowing": interesting adjacent tool, may matter later - "skip": no clear overlap Be OPINIONATED. Not neutral. Repo: vercel-labs/zerolang Stars: 3759 Language: C Topics: Description: The programming language for agents README (first 3000 chars): # Zero Zero is an experiment in building an agent-first programming language. The project is exploring what changes when agents are primary users from day one: a language that can be learned on the fly, tooling that exposes structured facts for debugging and repair, and a standard library broad enough that most programs do not start with a dependency search. Zero is pre-1 and intentionally unstable. The project will make breaking changes while it searches for the language, library, and tooling patterns that work best for agents. Treat today's syntax and APIs as something to explore, not something to memorize. If that sounds useful, try it with us: run examples, inspect the structured output, and send feedback about what helps agents work better. Security vulnerabilities should be expected. Zero is not ready for production systems, sensitive data, or trusted infrastructure. If you plan to run or develop Zero, do so in an isolated, disposable environment. ## What Zero Is Aiming For - Agent-first learnability: a small, regular language surface that agents can pick up quickly from examples, docs, and compiler feedback. - Standard-library depth: common capabilities should live in documented, coherent library APIs instead of scattered dependency stacks. - Deterministic tooling: diagnostics, graph facts, size reports, explanations, and fix plans should be structured enough for agents to inspect and act on. - Direct developer experience: checking, running, formatting, inspecting, and repairing code should be fast, copyable, and scriptable. - Regularity over syntax: prefer one obvious way to express most things, even when that makes code more explicit than a human might choose in another language. ## Quick Start Install the latest release: ```bash curl -fsSL https://zerolang.ai/install.sh | bash export PATH="$HOME/.zero/bin:$PATH" zero --version ``` Check a program: ```bash zero check examples/hello.0 ``` Run a small executable: ```bash zero run examples/add.0 ``` Expected output: ```text math works ``` ## Common Commands ```bash zero check examples/hello.0 zero run examples/add.0 zero build --emit exe --target linux-musl-x64 examples/add.0 --out .zero/out/add zero graph --json examples/systems-package zero size --json examples/point.0 zero skills get zero --full zero doctor --json ``` ## Validation ```bash pnpm run docs:test pnpm run conformance pnpm run native:test pnpm run command-contracts ``` Benchmarks run locally by default: ```bash pnpm run bench ``` Analyze this repo per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] { "what_it_is": "An experimental agent-first programming language from Vercel Labs designed to be learned and operated by AI agents with structured tooling output.", "stack": ["C", "TypeScript", "pnpm", "CLI tooling"], "why_mike_cares": "Direct overlap with Mike's agentic coding focus (Claude Code, Codex, OpenClaw fleet) and Vercel deployment stack, but it's a pre-1 language not a production tool.", "verdict": "worth-knowing", "verdict_reason": "Vercel betting on agent-first languages signals where agentic tooling is heading, but it's too early and unstable to use in Mike's active builds." }

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: The Next New Thing AI Video title: Anthropic beats OpenAI? + 5 AI Stories You Missed TRANSCRIPT (first 6000 chars): Anthropic is beating OpenAI in revenue, but that also means that they're taking money away from users like you and me. We got a solution for you. Codeex now can control your computer and one user had it fight Amazon and win him a refund. Grock is now enabled in Hermes agent and it can superpower your social media. We'll show you how. Google is coming out with a new laptop that's an AI first laptop. O one stupid thing about it, one useful thing. We'll talk about that, too. And finally, why are people putting their fingers in MacBooks when there's a terminal command for that? All that and so much more coming up in this week's news. Let's get into it. I think the best news story this week is that OpenAI's Codeex can control my computer. In fact, here this is from their demo. I'm going to show how they're using codeex and telling it to go and create a reminder in the reminders app because whether it's from your desktop app or the phone, you can now control apps. What do you think of this? So, I think it's a huge update, right? Computer use isn't a new thing, but they built it right into the Codeex app. And with codecs becoming a lot more popular, I mean, I feel like the vibe and kind of the sentiment in the community has shifted away from cloud code and toward codeex, there's going to be so many cool use cases that come out of this feature. And this is, you know, what we're about to see is one of probably a thousand plus use cases that you could come up with. >> I think so, too. My top use case is I can't get YouTube videos easily. um without using a browser and computer. I can't get um X data without using a browser and so I use a browser all the time. My big disappointment with all this is that I need to keep my computer going. I don't want to have a computer going. I want the cloud again. >> That seems to be a big pain point that people are having. I mean, before we hit record, we were looking at different like crazy ways that people were using to keep their computer running while the while their MacBook was closed. For example, if you've got like a Mac Mini like I do, I mean, obviously you can just set it to never sleep, but for folks that are vibe coding on MacBooks, there are workarounds, but they're a little janky. [ASSISTANT] { "tldr": [ "Codex (OpenAI) now has built-in computer use — one user had it wait on Amazon support for 25 minutes and negotiate a refund while he showered", "Community sentiment is shifting from Claude Code to Codex, partly driven by the computer-use integration", "Codex-as-chief-of-staff pattern: every 30 min check Slack/Gmail, prioritize, draft replies — drafts reportedly match user quality", "Keep-Mac-awake hack tax: amphetamine app or terminal `caffeinate` command needed because computer use requires the machine stay on (cloud execution still missing)", "Grok is now wired into Hermes agent for social media superpowering (mentioned, not detailed in this excerpt)" ], "tools": [ {"name": "Codex (OpenAI)", "url": "https://openai.com/codex", "description": "OpenAI's coding agent — now ships with computer-use built into the desktop/phone app"}, {"name": "Amphetamine", "url": "https://apps.apple.com/us/app/amphetamine/id937984704", "description": "Free macOS app to keep Mac awake — workaround for running Codex computer-use unattended"} ], "skill_candidates": [ {"slug": "codex-chief-of-staff", "description": "Recurring 30-min loop that scans Slack + Gmail, prioritizes unanswered threads, researches answers, and drafts replies in the user's voice"}, {"slug": "codex-computer-use-refund-agent", "description": "Pattern for dispatching Codex to wait in support chat queues and negotiate refunds/cancellations on behalf of the user"}, {"slug": "mac-keep-awake", "description": "Pre-flight check that enables caffeinate/amphetamine before launching long-running local-browser computer-use sessions on macOS"} ], "verdict": "worth-a-skim", "verdict_reason": "Codex computer-use is directly relevant to Mike's agentic-coding and automation domains and the chief-of-staff pattern is reusable, but the segment is a news roundup with shallow coverage and no novel tooling Mike couldn't replicate via Claude Code + browser agents he already pays for." }

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: Stephen G Pope Video title: I Stopped Paying For APIs I Can Run Myself TRANSCRIPT (first 6000 chars): Are you tired of wasting thousands of dollars on API subscriptions to support all of your automations? What if there was a free alternative? In this video, I'm going to show you how you can eliminate the costly monthly subscriptions for chat GPT whisper, Creatomate, JSON to video, cloud convert, and even tools like zero code kit that allow you to run Python code. And we'll replace it with one tool, the no code architects toolkit. Subscription services like OpenAI, Creatomate, and JSON Devideo run servers on the internet with some code that allows you to access their resources through an API for a monthly fee. And it's very easy for these subscription services to add up. And the No Code Architects toolkit is meant to replace all of these costly services. And how we help you do that is that we wrote some of our own code that you can easily deploy into your own server and access through an API inside of your own automations at almost no cost to you. Make sure to check out the documentation on GitHub. You can use chat GPT or feel free to jump into the noode architects community. Once you have this set up, if you're starting a scenario from scratch, you can come here to make, you can click HTTP. You can scroll down and make a request. Now for the URL, it's going to be the same URL that we copied here. Grab that. Come back and drop that into make. Now depending on which endpoint you are using, you're going to want to add the slash and then the full path of that endpoint URL. Let's head back to Postman. I'll head up here to concatenate. I'm going to grab this part of the URL here. Make sure you grab the slash and everything after it. And do not include that actual base URL. We actually copied that earlier right here. So, we don't need that. Put this into the URL box here. Make sure not to have two slashes here. I actually have two slashes, so I'm going to remove one. The method is a post. Next, we have to add the headers. And notice we're just setting the same settings here that we have here in Postman. Postman and the HTTP module essentially do the exact same thing. They just look a little bit different. So, we want to make sure we have that X API key from the headers and that password or the API key directly here in the headers here. X-a [ASSISTANT] {"tldr":["Stephen pitches the No Code Architects Toolkit as a self-hosted replacement for OpenAI Whisper, Creatomate, JSON2Video, CloudConvert, and ZeroCodeKit — all the paid APIs that pile up in Make/n8n stacks","Deploy the toolkit to your own server (Docker/Cloud Run), hit it from Make via HTTP module with X-API-Key header and JSON body — same pattern as Postman","Endpoints cover video concat, transcription, format conversion, and Python execution — Mike's pipelines (Creatify wraps, transcript pulls, video processing) could route through this instead of per-call API fees","Walkthrough demos /concatenate endpoint joining two 9s clips via POST → returns hosted URL ready to drop into next module","Lives on GitHub with docs + a community for setup help — pitched as the one-tool replacement for the entire SaaS automation stack"],"tools":[{"name":"No Code Architects Toolkit","url":"https://github.com/stephengpope/no-code-architects-toolkit","description":"Self-hosted API replacing Whisper/Creatomate/JSON2Video/CloudConvert/ZeroCodeKit for video, audio, code-exec endpoints"},{"name":"Make","url":"https://make.com","description":"Visual automation platform consuming the toolkit via HTTP module"},{"name":"Postman","url":"https://postman.com","description":"Used to prototype the API calls before porting to Make"}],"skill_candidates":[{"slug":"nca-toolkit-router","description":"Skill that routes video concat, transcription, format conversion, and Python exec tasks to a self-hosted NCA Toolkit deployment instead of paid APIs — replaces Creatomate/JSON2Video/Whisper calls in Mike's Creatify and transcript pipelines"},{"slug":"self-host-saas-api-replacement","description":"Pattern for auditing monthly SaaS API spend, identifying replaceable endpoints, and deploying open-source equivalents to a VPS with API-key auth"}],"verdict":"worth-a-skim","verdict_reason":"Tool is genuinely useful and overlaps with Mike's video/transcript automation spend, but it's a Make-focused walkthrough rather than a novel pattern, and Mike already self-hosts most of this stack via Python on his VPS."}

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: Nate Herk AI Automation Video title: What Karpathy Joining Anthropic Actually Means For Claude TRANSCRIPT (first 6000 chars): All right, so today is May 19th and a few hours ago, this tweet just went up. It was Andre Carpathy announcing that he has joined Enthropic. And if you don't know who that is, this is one of the most important people in modern AI. He was founding team at OpenAI. He ran AI at Tesla for like 5 years. He went back to OpenAI, left again, and then started an AI education company. And now he's anthropic. So normally the easy version of this video is, you know, big AI person joins big AI lab and that's the headline. But I think the more interesting question here is why Enthropic and why now? Because I'm assuming all of these big AI labs have wanted a guy like Cararpathy to hop on board. Because if you look at what Carpath has been building and talking about over the last multiple months and when you look at what Cloud Code has actually been shipping already, it almost starts to feel like these two things were kind of moving towards the same direction. So in this video, I want to show you the pattern that I think most people might miss when they sort of see this news, which is like the wrapper around the model, why your data and context are becoming the real product and where do I think Claude Code is probably heading next. So let's get into it. So real quick, just so that we're all on the same page, Andre Capathy is like one of the goats for sure. He's taken a really weird tour around the modern AI world because founding member at OpenAI back in like 2015. Then he ran AI at Tesla, came back to OpenAI in 2023, left again a year later, and then started Eureka Labs, which was basically his AI education company. And that's where he built stuff like LLM 101N, which is like a free course that teaches you how to build a language model from scratch. Not something that I know how to do. And he's also the person who like coined the term vibe coding, which is kind of what a lot of us are doing right now. You describe what you want in English and you let the AI write the code and then you kind of are just there vibing and steering it and editing and iterating with it. So he hasn't just been working in AI. He's been shaping how people think about and understand AI for a long time and now he's going to be at Enthropic and that ma [ASSISTANT] {"tldr":["Karpathy joined Anthropic on 2026-05-19, signaling Anthropic's bet on wrapper/context engineering over raw model leadership","Anthropic passed OpenAI in RAMP's business adoption index (34.4% vs 32.3%) — Claude Code is the wedge","New Anthropic enterprise services JV with Blackstone/Hellman/Goldman targets midsize ops embedding — the moat is application + adoption, not the model","Karpathy's 'context engineering' thesis maps 1:1 to Claude Code primitives: skills, subagents, hooks, MCP, CLAUDE.md, memory","Watch for Claude Code to double down on wrapper surface area (skills marketplace, deeper memory, IDE-native context) over the next 2 quarters"],"tools":[],"skill_candidates":[],"verdict":"worth-a-skim","verdict_reason":"Validates Mike's existing Claude Code + skills + MCP + CLAUDE.md investment thesis but introduces no new tools or extractable patterns — pure market signal."}

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: Leon van Zyl Video title: Claude Code Builds Portfolio Websites That Look Like This TRANSCRIPT (first 6000 chars): A few weeks ago, I redesigned my personal portfolio page to look and feel like a coding agent CLI, like Claude Code. Complete with tool calls, response streaming, it looks like a user interacting with the agent asking questions about myself. And as an added bonus, users can even send it questions and it will answer questions from a knowledge base. How's it? I'm Leon. Whenever I share my portfolio page, people ask me how I build it. In today's video, I'll build this with you. We'll start with a blank project, use Claude Code to build this and deploy to production within a few minutes. And by the way, feel free to check out my website. There are tons of Easter eggs in this. From keyboard shortcuts to slash commands. There's a lot of fun stuff in here. Now for the fun part, let's build this. Open up a blank folder and in the terminal run the command npx create next app at latest space period. This will download and install Next.js. Since we're using Next.js, I like to download and install the next best practices skill. So simply copy this command, then back in our project's terminal, let's paste in that command, then select Claude Code, press enter, install that project level, and it's complete the installation. And secondly, I'm also going to install the front end design skill. This will give us a way better UI experience. Right, let's open our coding agent. I'll be using Claude Code in this video, but it really doesn't matter. Use whatever coding agent you're comfortable with. Then let's switch to pattern mode and to save on time, I'm simply going to paste in the prompt. By the way, you can download these prompts and the code absolutely free. I'll link to the Knob repo in the description. Also, to create this new folder called resources and in that is create a folder called prompts. And in that folder, you'll find this research and port prompt. It's very simple. It simply tells the agent to build a terminal style page. This is going to tell the agent to do some research on you. So you can provide links to your LinkedIn, your YouTube channel X. You can even upload a CV if you want. Another agent will go through all of these resources to do some research on you and create this page. So back in Claude Code, it's based in th [ASSISTANT] { "tldr": [ "Leon builds a Claude Code CLI-styled portfolio site in Next.js with streaming tool-call animations, fake terminal UI, and a regex-matched Q&A chatbot fed by a qa.json file.", "Workflow: npx create-next-app, install Next.js best-practices skill + frontend-design skill into Claude Code, then feed 3 sequential prompts (research/port, streaming animations, Q&A chat) from a public GitHub repo.", "Animations play once per session so repeat visitors don't have to rewatch; a 'new session' button replays the stream.", "Q&A is not real inference, it's regex pattern matching against qa.json, useful as a zero-cost portfolio gimmick.", "Deploys to production via git commit + publish branch (Vercel-style flow) in minutes." ], "tools": [ {"name": "Claude Code", "url": "https://www.anthropic.com/claude-code", "description": "Coding agent used to build and iterate the portfolio site"}, {"name": "Next.js (create-next-app)", "url": "https://nextjs.org", "description": "Framework scaffolded via npx create-next-app@latest"} ], "skill_candidates": [ {"slug": "cli-style-portfolio-builder", "description": "Three-prompt sequence (research+port, streaming tool-call animations, regex Q&A) to generate a Claude-Code-CLI-styled personal portfolio in Next.js with one-shot session animations and qa.json chat"}, {"slug": "fake-streaming-tool-call-ui", "description": "Reusable Next.js component pattern that simulates agent tool calls and response streaming for marketing/portfolio pages, with per-session replay gating"} ], "verdict": "worth-a-skim", "verdict_reason": "Cute aesthetic pattern and a clean three-prompt workflow Mike could lift for Merlino AI's own site, but no new tools and the Q&A is fake regex, not a novel agent pattern." }

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: Julian Goldie SEO Video title: Claude AI SEO is INSANE! TRANSCRIPT (first 6000 chars): Today, I'm going to show you how to rank number one with Claude AI SEO, and this is a really powerful system for ranking your website. It's also the fastest and the easiest way I know to rank your website with AI SEO. And basically, what we have here is a full system. I'm going to walk you through five different methods that you can use within this whole system to rank. And I'll show you proof that every single one works as well, right? So, for example, you can see this SEO machine that we've got right here with Claude. And what it can do is actually deploy websites and transcripts and, for example, blog posts as well, and then get them ranked directly on Google, which is a super powerful system, right? And so, you can see the actual post right here. If you want to see proof that this actually works, here's some examples. So, if we pull up these websites, I'm going to show you some examples of our websites actually ranking with these methods. So, you can see, for example, this website went from like zero to a peak of 71 clicks per day from Google directly, right? Here's another one. So, this website went from like basically nothing, like nine clicks a day, all the way up to a peak of 1,134 clicks per day. If we have a look at this website, it went from like four clicks a day all the way up to 243 clicks a day. It's ranking on Google pretty nicely. And this one follows a similar sort of trajectory, right? Four clicks a day all the way up to a peak of 95 clicks per day, right? Four different websites all ranking all ranking really nicely, as you can see right here. So, it's a really powerful system that I'm going to show you, and you can do all inside this beautiful user interface, as you can see. I mean, this is probably the most powerful dashboard I've ever seen when it comes to using AI SEO. Like, literally, all you do is you plug in your target keyword, and then you add in a case study. From there, you're going to deploy the websites and get them ranking pretty quickly as well, right? And there's five different methods. This is just one. There's so much more. This This really, really deep, which I'm excited to show you. So, let's get straight into this. If you want to see some examples of us ranking, let's have a look here. So, you know, here's [ASSISTANT] { "tldr": [ "Julian pitches a 'Goldie Search Gravity Stack' / 'Agent OS' dashboard that takes one keyword + case study and auto-deploys ranking websites, blog posts, and AI-search content.", "Shows 4 GSC screenshots (peaks of 71, 1134, 243, 95 clicks/day) as proof, plus rankings in Google AI Mode and Perplexity for branded queries.", "Five-layer system anchored on a 'content engine' that replaces the manual ChatGPT to Claude to WordPress to Ahrefs workflow with a single UI.", "Heavy infomercial framing for Julian's paid product, no actual Claude prompts, MCP configs, or repo links shared in the transcript window.", "Core idea Mike already runs at higher fidelity: keyword to AI content to multi-surface ranking (Google + AI engines) via Einstein/Shakespeare/Tommy." ], "tools": [], "skill_candidates": [], "verdict": "skip", "verdict_reason": "Sales pitch for Julian's gated 'Agent OS' product with screenshot proof but zero shared mechanics, no tool URLs, and nothing Mike's existing SEO agent fleet doesn't already do better." }

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: Julian Goldie SEO Video title: China's New Qwen 3.7 is INSANE! TRANSCRIPT (first 6000 chars): China's new Qwen 3.7 is insane. What if I told you the AI model everyone was just hyping up is already old news? What if China just dropped something so wild it's making Silicon Valley nervous all over again? What if the next big AI leap didn't come from OpenAI, Google, or Anthropic? It came from Alibaba, and it just landed on the leaderboards out of nowhere. Hey, I'm the digital avatar of Julian Goldie, and I help you actually use AI tools to get real work done. In the next few minutes, I'm going to show you exactly what Qwen 3.7 is, what it can already do, and the one thing you should be doing right now to stay ahead of the curve. Trust me, by the end of this, you're going to look at your AI stack completely differently. So, here's what's going on. Qwen team over at Alibaba just dropped two brand new preview models on Arena, the platform that used to be called LM Arena. They're called Qwen 3.7 Max Preview and Qwen 3.7 Plus Preview, and they didn't just show up, they climbed. Qwen 3.7 Max Preview landed at number 13 overall in Text Arena, which bumped Alibaba up to the number six lab in the world for text. Here's the part that's really turning heads. Vision Arena, where models go head-to-head on understanding images, Qwen 3.7 Plus Preview ranked number 16 overall. That puts Alibaba at number five worldwide for vision. Number five? For an open weights-focused lab from China, that is wild. And in the expert arena, the toughest one where only expert-level prompts are judged, Qwen 3.7 Max Preview came in at number nine. In the subcategories, it ranked seventh in math, ninth in software and IT, and tenth in coding. These are top 10 finishes across the board. Remember, this is just a preview. The full release hasn't even dropped yet. So, what exactly is Qwen 3.7? Qwen is Alibaba's family of large language models. Think of them like Alibaba's version of GPT or Claude. The team has been releasing them in waves. Qwen 3.5 came out in February. Qwen 3.6 Plus dropped in early April, and now just weeks later, we're already getting Qwen 3.7 previews. That's how fast things are moving. Max is the flagship, built for top-tier performance on the hardest tasks. Plus is the more balanced option. Right now, in preview, both only support thinking mode, which i [ASSISTANT] {"tldr":["Alibaba dropped Qwen 3.7 Max Preview and Plus Preview on Arena — Max hit #13 overall in text, #9 in expert, top-10 in coding/math; Plus hit #5 globally in vision","Both previews are thinking-mode only right now (no tools/search/code interpreter yet); official launch tied to Alibaba Cloud summit on May 20","Qwen 3.6 Plus already shipped a 1M-token context window, so 3.7 is a credible repo-level coding contender at China-tier pricing","Julian's demo: one-shot HTML SEO ROI calculator (traffic, conv rate, AOV, lift) built end-to-end in a single prompt","Rest of the video is a soft pitch into Julian's AI Profit Boardroom — the model news is the only real signal"],"tools":[{"name":"Qwen Chat","url":"https://chat.qwen.ai","description":"Alibaba's chat interface where Qwen 3.7 Max/Plus Preview are live"},{"name":"LMArena (formerly LM Arena)","url":"https://lmarena.ai","description":"Blind side-by-side model leaderboard where Qwen 3.7 previews are ranking"}],"skill_candidates":[{"slug":"qwen-coder-router","description":"Route long-context repo-level coding tasks (>200k tokens) to Qwen 3 Max via OpenRouter/DashScope as a cheaper alternative to Claude/GPT for bulk refactors and codebase reads"}],"verdict":"worth-a-skim","verdict_reason":"Legit model news worth knowing (Qwen 3.7 previews + vision top-5 + 1M context for coding) but the video is mostly hype and a Boardroom upsell with no novel pattern or workflow Mike doesn't already have."}

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: Julian Goldie SEO Video title: Claude + Codex Is INSANE! TRANSCRIPT (first 6000 chars): Claude plus Codex is insane. Right now, and the gap between these two AI tools and everything else just got way bigger. I'll show you why in a second. But first, here's the thing most people are missing. This isn't a normal update. Claude and Codex are no longer just coding helpers. They're full agent NC workspaces. They run tasks. They open sessions. They work across your apps. They touch your whole repo. They keep going when you close the laptop. A year ago, these tools could write a function for you. Maybe finish a line of code. Now they ship features while you sleep. And both companies, Anthropic and OpenAI, are racing toward the exact same future. They work like teammates. Not autocomplete. Not chatbots. Teammates that handle real work end to end. Let me show you what's actually going on. Because once you see this, you'll get why so many business owners are quietly moving fast right now. Start with Claude. Anthropic just keeps shipping. Claude code is the command line tool. Lives in your terminal. You give it a task, it goes and does the task. It plans, runs commands, edits files, it tests its work, it fixes its own mistakes, and it comes back and tells you what it did. That alone is wild. They didn't stop there. They added something called background tasks. You hand Claude a job. You close the window. Claude keeps working. You come back later and the job is done. Think about that for 1 second. Your AI is working when you're not. They added webhooks. And Claude can be triggered by other tools. New email comes in. New lead fills out a form. A new file lands in your drive. Claude kicks off. It does the work. Reports back. They added multi-agent setups. And Claude is the boss. The other Claudes are the workers. The boss splits the job. The workers run in parallel. And the boss puts it all together. They added stronger memory. Claude can now hold on to project notes, your style, your rules, your past work. You stop repeating yourself every single time. Now flip over to Codex. But I just pushed Codex into the same lane. Codex used to be a model. Codex is a workspace. You get the CLI. You get the IDE plugin. You get the cloud version. Get the web app. All of them share the same brain. All of them share the same tasks. All of them can hand work bac [ASSISTANT] {"tldr":["Claude Code and Codex have shifted from autocomplete to full agentic workspaces — background tasks, webhooks, sub-agents, GitHub PR automation","Claude background tasks let you hand off a job, close the laptop, and have it ship while you sleep — Mike's already doing this with his fleet","Codex now spans CLI + IDE + cloud + web app sharing one brain, opens PRs with tests and descriptions auto-written","Sub-agent pattern (research/writer/tester/reviewer) validated by Anthropic — matches Mike's Oliver/Carlos/leads orchestra model","Anthropic internal teams shipping sprint-sized features in an afternoon; Codex cloud cutting days to hours"],"tools":[{"name":"Claude Code","url":"https://claude.com/claude-code","description":"Anthropic's agentic CLI with background tasks, webhooks, sub-agents, memory"},{"name":"Codex","url":"https://openai.com/codex","description":"OpenAI's agent workspace across CLI, IDE, cloud, web with shared brain and GitHub PR automation"}],"skill_candidates":[],"verdict":"skip","verdict_reason":"Surface-level rehash of Claude Code + Codex features Mike already runs daily — no new tools, no novel pattern, half the video is a paid community pitch."}

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: Julian Goldie SEO Video title: NEW Google Gemini 3.5 Flash is WILD! TRANSCRIPT (first 6000 chars): Google just dropped Gemini 3.5 Flash and this changes everything. This is the most powerful AI agent update Google has ever shipped. And today I'm going to show you exactly how to use it to automate your work, build full websites, and run an entire team of AI agents all at the same time. Now, here's what makes this different from anything else out there. This model is four times faster than any other Frontier AI. It's smarter than Google's previous pro model, and it's free for everyone to use. I'll show you how today. And by the end of this video, you're going to have Google Gemini 3.5 Flash wired directly into your AI agents working for you 24/7. There's also one thing I'm going to show you towards the end that most people are completely missing with this update, and it's a thing that makes this 10 times more powerful. Today, we have a brand new update from Google Gemini 3.5 Flash, and I'm going to show you exactly how to use it. You can see for example, we actually built out this full blog and website using Google Gemini 3.5 Flash. All the images were generated with AI. All the copy, all the design of the website and even the deployment was handled with AI. And this is a super powerful system. If you know how to use it, which is what I'm going to show you today. If you have any questions as you go along, feel free to ask. And what we've got here is the newest, latest version of Flash, which is crazy fast, right? Let me give you an example. So you can see here 3.5 flash. That's what we've selected. This literally just dropped hours ago from Google. For example, if we're like, okay, are you working? Just to check like the speed of it, it's super fast to reply, right? Which is really cool. You can see here it just it replies super fast. And if you're wondering, okay, how does this work? What does it mean? What the difference is? I'm going to run you through all of that today. So this is the new announcement just dropped. So it's Gemini 3.5. As you can see, this is available now inside Gemini directly. And we've actually built it into agent OS which you can see right here. So we actually have Gemini, Anti-gravity, Hermes, OpenC Claw, and Claude all working side by side inside this dashboard. Right. And this is pretty amazing because with [ASSISTANT] { "tldr": [ "Julian hypes a 'Gemini 3.5 Flash' release but the actual current Google model is Gemini 2.5 Flash — treat the model name as suspect or rebranded clickbait.", "Real signal: Julian is running a multi-agent 'Agent OS' dashboard with Gemini, Anti-gravity 2.0, Hermes, OpenClaw, and Claude side-by-side — same orchestrator pattern Mike already runs.", "Claimed benchmarks: 76 terminal bench (vs 58 on prior Flash), 83.6% on agentic tasks (beats Gemini 3 Pro at 78%) — if true, Flash becomes the cheap default for agentic loops.", "Anti-gravity 2.0 CLI can be plugged into a mission control dashboard to scaffold + deploy full websites end-to-end.", "Gemini Spark (new 24/7 personal agent) is US-only on Google AI Ultra plan — gated, not immediately usable." ], "tools": [], "skill_candidates": [], "verdict": "skip", "verdict_reason": "Surface-level hype video with a likely-wrong model name, no novel tools, no actionable workflow Mike doesn't already have in his own multi-agent dashboard." }

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: Julian Goldie SEO Video title: Hermes + Antigravity 2.0: Build ANYTHING! TRANSCRIPT (first 6000 chars): Imagine having Hermes agent and anti-gravity working together in one place, building, automating, and creating for you 24/7. I just ran a full masterclass breaking down exactly that, and in this video I'm going to show you the exact setup, how to combine Hermes and anti-gravity, so your agents know who you are, understand you, and get smarter every single day. I'm also answering the biggest questions people have right now about context limits, model choices, migrating from other tools, and what actually works in the real world. If you've ever felt stuck or overwhelmed, or like you're not getting the results you should be getting with AI, this is a video that fixes that. Let's get into it. What I'm going to do is just answer some questions inside our community, and let's see what we got. Ah, so this is interesting. So, for example, Alan has just posted inside the AI Profit Bot in about anti-gravity, and he says be careful with anti-gravity 2 cuz it's a major change from the old version, and Google removed the built-in Visual Studio Code style editor and terminal. So, it's now a standalone agent only app, designed purely for chatting with agents managing projects. So, you can actually use this guide to go back step-by-step. That's why I've set up the Agent Operating System really, because if you look at this, it's more agentic, the apps, right? Anti-gravity just it is like a standard chat, as you can see, but if you want to turn it into something fully agentic and work with all your AI agents, then you can use the OS instead. What else you got here? Kenny says, "Can anyone suggest like an AI agent chat workflow with NA10?" We actually got training on this over here. So, we actually have an example tutorial on how to do that right here, to build your first sort of chat agent essentially. Let's see what questions we got here. Sumit says, "Anti-gravity 2 with context 7 consider more powerful." Honestly, I think the best thing you can do is just give it more context about you, who you are, etc., and then it'll be way more agentic, right? That's what we've done with the memory system here, and all the skills that we have on the left-hand side. For example, like the SEO tool, the goals section, the journal, the Kanban board, the notebook, [ASSISTANT] { "tldr": [ "Google stripped the VS Code-style IDE and terminal from Antigravity 2.0, turning it into a chat-only agent app (breaking change from v1).", "Julian's pitch: bolt Antigravity 2.0 onto his 'Hermes Agent Operating System' to recover the agentic workflow with memory, skills, Kanban, journal, etc.", "He claims Hermes replaces n8n for most automation jobs: computer-use setup went from 3 hours in n8n to 20 minutes in Hermes.", "Recommended low-resource stack: MacBook + OpenClaw + Hermes + Owlpha API (free tier).", "OpenClaw + WordPress login flow is the recommended path for automated SEO content deployment; Hyperframes or Remotion for video." ], "tools": [], "skill_candidates": [], "verdict": "skip", "verdict_reason": "Affiliate/community-promo Q&A pitching Julian's paid 'Hermes Agent OS' with no concrete tutorials, code, or novel patterns; the one real signal (Antigravity 2.0 dropped its IDE) is a 10-second fact Mike doesn't need a 20-min video for." }

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: Julian Goldie SEO Video title: Hermes Agent OS is INSANE! 🤯 TRANSCRIPT (first 6000 chars): Hermes Agent OS is insane. I gave an AI agent one sentence, it built a website, ranked it on Google, and made a video. This is Hermes Agent OS, and it might be the most powerful free automation tool available right now. I typed one sentence, it wrote the content, deployed a full website to a live domain, and ranked it on the first page of Google. No coding, no Zapier, no n8n, just one AI agent doing all the heavy lifting. Need videos? Connect it to Hyperframes or Remotion, and it edits and generates them automatically. Need a smarter agent? Link your Obsidian vault, and it learns who you are, what you do, and gets better single day. The old way took hours of setup, broken APIs, and technical headaches. The new way, one sentence, done. I can't believe this is free. Extract the structured JSON signal per your instructions. Return ONLY valid JSON, no markdown fences. [ASSISTANT] {"tldr":["Hermes Agent OS demoed as a one-sentence-to-deployed-website agent that allegedly handles content, deploy, and ranking end-to-end","Claims integration with Hyperframes/Remotion for automated video generation and Obsidian vault linking for personalized agent memory","Pitched as a free no-code/no-Zapier/no-n8n alternative for full-stack automation","Heavy hype framing with zero technical detail, no repo link, no architecture, no proof of ranking shown","Julian Goldie content pattern: tool shilling with affiliate intent, treat claims as unverified"],"tools":[{"name":"Hermes Agent OS","url":"","description":"Claimed one-sentence-to-deployed-site AI agent with video and Obsidian integrations"},{"name":"Remotion","url":"https://www.remotion.dev","description":"Programmatic video generation in React, already on Mike's radar"},{"name":"Obsidian","url":"https://obsidian.md","description":"Markdown vault used as agent memory source"}],"skill_candidates":[],"verdict":"skip","verdict_reason":"Julian Goldie hype reel with no technical substance, no verifiable Hermes Agent OS URL, and nothing Mike's existing Oliver/Carlos/Hermes stack doesn't already do better."}

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: Julian Goldie SEO Video title: NEW Antigravity 2.0 + Agent OS is INSANE! TRANSCRIPT (first 6000 chars): Google Anti-Gravity 2.0 just dropped and your AI agents just became 10 times more powerful. This is Google's most advanced agent platform ever built and today I'm going to show you how to plug it into a full command center so you have an entire team of AI agents building websites, creating content, automating your work all at the same time. Now, here's what makes this different. Most people open up Anti-Gravity, they type prompt, they get average results because they're missing this one thing and I'm going to show you exactly what it is today. And when you set this up the right way, your know who you are, they know your business and they never start from zero again. So, every single day they get smarter, every single output gets better and every single day your AI operation improves over time. So, by the end of this video, you're going to have Anti-Gravity 2.0 running inside a full agent operating system, more powerful, more organized, more automated than anything you've ever seen before. Stick with me to the end because there's one upgrade I'm going to show you that makes everything click. Today, we have a brand new update from Google which is Google Anti-Gravity 2.0. So, Google Anti-Gravity 2.0 is a dedicated platform to work with AI agents. You can orchestrate multiple autonomous AI agents working in parallel across independent tasks. Basically, what that means is you can use this, it looks very similar to Codex now. They've totally changed the UI. Basically, this allows you to create whatever you want. So, you can build out websites, you can build out teams of agents, you can have it organized for websites for you. So, you can see for example, I built out this full website right here using Anti-Gravity. And with Anti-Gravity 2.0, it's also got Gemini 3.5 Flash plugged in which is super powerful stuff, right? Now, also on top of that, the new updates include dynamic sub-agents. So, sub-agents is defined basically is like a team of agents that work together, right? Then you've also got scheduled tasks. So, you'll see over here you have scheduled tasks and you can give it new tasks to schedule and automate routine checks which is pretty cool. You have artifacts. So, these are deliverables from the agent to communicate its progres [ASSISTANT] {"tldr":["Google Antigravity 2.0 launched as a dedicated agent platform with Gemini 3.5 Flash, dynamic sub-agents, scheduled tasks, artifacts, and MCP support","Gemini CLI is being rebranded to Antigravity CLI, giving terminal access similar to Claude Code","Julian wraps Antigravity CLI inside an 'Agent OS' mission control dashboard alongside Claude, Hermes, and Gemini for unified memory and orchestration","Memory layer is tied to an Obsidian vault so agents inherit persistent context about the operator and projects","Kanban-style multi-agent assignment with NotebookLM, image/video/speech studio bolted into the same workspace"],"tools":[{"name":"Google Antigravity 2.0","url":"https://antigravity.google","description":"Google's agent orchestration platform with sub-agents, scheduled tasks, MCP support, and Gemini 3.5 Flash"},{"name":"Antigravity CLI","url":"https://antigravity.google/cli","description":"Rebranded Gemini CLI for terminal-based agent control, comparable to Claude Code"}],"skill_candidates":[{"slug":"antigravity-cli-orchestration","description":"Plug Antigravity CLI into a multi-agent mission control dashboard alongside Claude Code, Gemini, and Hermes with shared memory routing"},{"slug":"agent-os-memory-bridge","description":"Pattern for binding agent CLIs (Antigravity, Gemini, Claude) to a persistent Obsidian-backed memory vault so every agent inherits operator context"}],"verdict":"worth-a-skim","verdict_reason":"Antigravity 2.0 + CLI rebrand is a legit new tool hitting Mike's agentic-coding and MCP domains, but the video is mostly a dashboard demo with no deep implementation detail."}

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: Julian Goldie SEO Video title: NEW Google Updates Just Changed AI FOREVER! TRANSCRIPT (first 6000 chars): Google IO 2026 just happened and I'm going to be real with you. Google IO just dropped some of the biggest AI updates we've seen all year and I've been watching these announcements for a long time. I know when something is just noise and I know when something is actually different. This one is different because Google didn't just release a faster model. They released an entirely new way of running your business with AI. Agents will work 24/7 in the background. a model that is four times faster than anything else at the frontier with Google and smarter than the previous best model and a desktop app that lets you run a full team of AI agents in parallel plus an AI agent called Gemini Spark that keeps working after you close your laptop. We're going to go through every major announcement from Google IO 2026. I'm going to show you what each one actually means for your business. And I'm going to show you the one thing most people are completely missing, which is how to wire all of this into one command center so that it improves and actually helps you grow as you grow every single day with this stuff. Let's start with the numbers because Sunda Pichai opened with this and the numbers alone tell you everything about where this is going. Two years ago, Google was processing 9.7 trillion tokens per month. Last year at IIO that number was 480 trillion. Today in 2026 over 3.2 quadrillion tokens per month. That's a 7x jump in one single year. 8.5 million developers are now building apps with Google's models every single month. Their APIs are processing 19 billion tokens per minute. The Gemini app went from 400 million monthly active users last year to 900 million today. an AI mode. Google's upgrade to search already hit 1 billion monthly active users in under a year. So, these are not projections. These are current live numbers. AI adoption isn't slowing down. It's accelerating faster than any point before. And everything Google just dropped on iOS designed to pour more fuel on that fire. So, let's get into the actual drops. Starting with the model that powers everything else, and that is Gemini 3.5 Flash. So, this is Google's new flagship, and the headline numbers are genuinely hard to ignore. four times faster than any other Frontier mode [ASSISTANT] {"tldr":["Google IO 2026 dropped Gemini 3.5 Flash: 4x faster than frontier models, beats Gemini 3.1 Pro on coding/agentic benchmarks (83.6% vs 78%), costs less than half","Antigravity 2.0 ships as a standalone desktop app for running multiple AI agents in parallel — demo had 93 sub-agents build a working OS with 2.6B tokens","Gemini Spark introduced as a background agent that keeps working after you close your laptop","Google processing 3.2 quadrillion tokens/month (7x YoY), Gemini app at 900M MAU, AI Mode in search hit 1B MAU in under a year","Julian pitches his paid 'AR Profit Boardroom' agent OS command center as the wrapper around all this — skip the pitch, take the tool names"],"tools":[{"name":"Gemini 3.5 Flash","url":"https://deepmind.google/technologies/gemini/","description":"Google's new flagship flash model, 4x faster than frontier, beats prior Pro on agentic benchmarks at sub-half cost"},{"name":"Antigravity 2.0","url":"https://antigravity.google/","description":"Standalone desktop app for running AI agents in parallel with a unified command center"},{"name":"Gemini Spark","url":"https://gemini.google.com/","description":"Background AI agent that continues work after the user closes their laptop"}],"skill_candidates":[{"slug":"antigravity-agent-orchestrator","description":"Wire Antigravity 2.0 into Mike's existing fleet (Claude Code, OpenClaw, Hermes) as another parallel-agent execution lane alongside Carlos"},{"slug":"gemini-flash-router","description":"Cost-routing rule that shifts bulk/cheap agent workloads to Gemini 3.5 Flash when Opus-class reasoning isn't required, mirroring Mike's existing model-tier strategy"}],"verdict":"worth-a-skim","verdict_reason":"Hits agentic coding + LLM tooling + parallel agents (3 Mike domains) with real new releases (Gemini 3.5 Flash, Antigravity 2.0, Gemini Spark), but the video is mostly a sales funnel for a paid boardroom — extract the announcements and skip the rest."}

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: Julian Goldie SEO Video title: NEW Google Stitch 2.0 Updates Just Changed Everything! 🤯 TRANSCRIPT (first 6000 chars): New Google Stitch 2.0 updates just changed everything. Google Stitch 2.0 just changed how anyone builds an app or a website, and you don't need to be a designer to use it. I want to show you something that happened today. Google held their big IO event, and they updated a free tool called Stitch. Watch what it does now, because this is wild for anyone who runs a business. Here's the old way to design a website. You hire a designer. You wait 3 or 4 days. Send you a file. You ask [music] for changes. You wait again. By the time it's done, a week is gone. Here's the new way with Stitch 2.0. You type what you want. You just talk to it, and it builds a design right in front of you, live while you watch. No waiting. No designer. No file sitting in your inbox. Let me say that again, so it lands. Talk. Build. You watch it happen on the screen in real time. That's the whole point of this update. Now, what is Stitch? Quick version. It's a free tool from Google. You go to a website, you sign in with a Google account, and you tell it what kind of app or web page you want. It makes it for you. Real design, real buttons, real layout. Not a rough sketch. Stitch came out last year, but it could only do one thing at a time. You type, it would make one screen, and that was it. Slow. Clunky. Today's update fixed all of that. So, let me walk you through the five big changes, and one bonus that I think is the best part. The first change is streaming. This is the one that feels like magic. Before, you type your idea and wait. Now, Stitch shows its work as it goes. You watch the page get built piece by piece right on the screen. It streams its work straight to the canvas, and you can see it working in real time. Why does that matter to you? Because you can stop it early. If it's going the wrong way, you jump in and steer it before it finishes. The Stitch agent lets you steer the design before the final product is done. You're not stuck waiting for a bad result. You fix it as it happens. Here's a simple way you'd use this. Say you want a fresh sign-up page for a community of business owners learning AI. You watch Stitch build it. Halfway through, you say, "Make it feel more premium." And it shifts right there. The second change is starting [ASSISTANT] { "tldr": [ "Google Stitch 2.0 (free, Google account) now streams designs live to canvas so you can steer mid-build instead of waiting for a finished bad result", "In-place edits finally work: click text/images/spacing directly on canvas, no re-prompt loop", "Import existing assets (screenshots, code, copy) as a starting point so Stitch matches your existing brand instead of generating from zero", "Real exit doors: Figma export with clean auto-layouts, Tailwind+HTML code export, one-click Netlify publish, handoff to Google Antigravity for backend", "Interactive play mode simulates real navigation across linked screens (hover states, inputs, multi-screen flows) so you catch UX issues before build" ], "tools": [ {"name": "Google Stitch", "url": "https://stitch.withgoogle.com", "description": "Free Google AI design tool, prompt to UI with live streaming, Figma/Tailwind/Netlify export"}, {"name": "Google Antigravity", "url": "https://antigravity.google", "description": "Google's agentic coding environment for wiring backend logic into Stitch designs"}, {"name": "Netlify", "url": "https://netlify.com", "description": "One-click publish target now wired directly into Stitch"} ], "skill_candidates": [ {"slug": "stitch-to-nextjs-pipeline", "description": "Prompt Stitch 2.0 to design a page, export Tailwind+HTML, port into Next.js/ShadCN project, swap stock placeholders for brand assets"}, {"slug": "stitch-screenshot-clone", "description": "Feed an existing client page screenshot into Stitch as the starting point to generate matching-brand variants for location/service pages at scale"} ], "verdict": "worth-a-skim", "verdict_reason": "Stitch 2.0's Tailwind export + Netlify publish + screenshot-to-design hits Mike's Next.js/ShadCN/Vercel stack and multi-brand site work, but it's a UI-only tool with no agent/MCP angle and the video is a surface-level feature recap." }

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: Jeff Su Video title: Top 5 Claude Cowork Tips I Wish I Knew from Day One TRANSCRIPT (first 6000 chars): Co-work is insanely powerful, but there's a problem. Right now, there's no gold standard on how to set up your workspace. So, if you get the foundation wrong, you're going to keep running into avoidable issues down the line. So, after 5 months of using co-work daily to run my entire life and going to debt to pay for token usage, here are five essential things to get right from day one. Let's get started. Kicking things off with tip number one, the markdown translator. As you know by now, Co-work's instructions and memory live in these MD markdown files. And although we can open them up and edit this directly, opening this just costs 20 tokens. Okay, that's a joke, but I kind of feel it's true. It's a pain to read like this, right? And annoying as hell to edit. So, first what you want to do is to install a free app called Obsidian. Open folder as vault. Open. Point it to your co-workspace folder. Open. And now every MD file instantly renders with proper headings, bold text, and bullet points. Basically, a much more readable format. And now let's say I want to change something in this claw.md file. Instead of doing anything here, I can select the claw.md tab in Obsidian and replace this first bullet point. For example, under preferences with always make inappropriate jokes. And let's just remove that line from earlier. And I'm going to close this and reopen. And you will see that the changes are already there. To be clear, you don't need to learn Obsidian or use any of its other features. It's just a lens to read and edit MD files. Pro tip, you can click command and control plus to zoom in. You can click the reading mode icon to lock the Obsidian page so you don't make edits by mistake. And you can even go to Obsidian settings, files and links. Keep the show all file types toggle turned on. And this lets you see non.mmd files like spreadsheets, PDFs, and even images in the sidebar. Moving on to tip number two, the 300 line rule. Because a root cloud.md loads every single session, a bloated file wastes a lot of tokens. And when I cut mine from over 600 lines to around 250, my token usage dropped by roughly 25%. And here are three tactics you can use right away. First, only include the bare essentials. My claw.md template has six sect [ASSISTANT] { "tldr": [ "Install Obsidian and point it at your Claude Code workspace folder so every CLAUDE.md and memory.md renders as readable markdown instead of raw text you waste tokens editing", "Hard cap your root CLAUDE.md at 200-250 lines (300 absolute max) - Jeff cut his from 600 to 250 and dropped token usage ~25%", "Use a 6-section CLAUDE.md template: memory system, preferences, rules, routing map, references (on-demand pointers), and workstation-creation guide", "Relocate task-specific rules out of root CLAUDE.md into reference files loaded on-demand - ask Claude to do the move and replace with a one-line pointer", "Split prescriptive content (always/never rules) into CLAUDE.md vs descriptive memory (what happened, decisions made) into memory.md - mixing them tanks output quality" ], "tools": [ {"name": "Obsidian", "url": "https://obsidian.md", "description": "Free markdown editor used as a lens over the Claude Code workspace folder for readable CLAUDE.md/memory.md editing"} ], "skill_candidates": [ {"slug": "claude-md-diet", "description": "Audit a bloated root CLAUDE.md, identify task-specific sections, relocate them to on-demand reference files, and replace with one-line pointers - target 200-250 lines"}, {"slug": "claude-workspace-scaffold", "description": "Bootstrap a Claude Code workspace with the 6-section template (memory system, preferences, rules, routing map, references, workstation creator) plus Obsidian vault config"}, {"slug": "memory-vs-rules-router", "description": "Classify each line of agent instructions as prescriptive (CLAUDE.md - always/never) vs descriptive (memory.md - what happened) and route to the correct file"} ], "verdict": "dont-miss", "verdict_reason": "Directly hits Claude Code + skills + prompt engineering domains with a concrete token-saving pattern (25% reduction) and a CLAUDE.md hygiene methodology Mike can apply to his own bloated global CLAUDE.md immediately." }

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: Income stream surfers Video title: Gemini 3.5 Flash + Antigravity 2.0 = INSANE Coding (FREE) TRANSCRIPT (first 6000 chars): Okay guys, so I'm going to break down everything that's happened at the Google IO so far. This video is before they announced their new flagship model. This video is more about the kind of things that they announced before the new flagship model. Let's just jump into things. Okay guys, so if you weren't watching the Google IO, then it's happening right now. And the main thing that they have announced so far is Gemini 3.5 Flash. Now for me personally, this is absolutely massive. I've been saying that I wanted a new Gemini flash model for ages, right? I've been using or I did use for a very, very long time Gemini 3 flash inside anti-gravity and it was probably the best experience I've ever had with Google AI systems. So, this is what they've released. It's an incredibly fast new model. I was pretty excited about this. As you can see, I have been waiting for this specifically and I've been I've said this a million times. I really, really wanted a new Gemini Flash model. That's what we got today. It's not out just yet on anti-gravity. I've been sat here trying to check for updates every 10 minutes. It's still not available for me. It'll probably be available in America first and then eventually it'll drift over to Europe. Um, intelligence versus output. It's like one of the fastest models. 300 tokens per second for extremely good intelligence. Apparently, they built an OS system, which is like the classic uh test. It's the one that they also did to show the power of I think it was Opus 4.7 at Anthropic, and it basically created an entire OS for $1,000, which is absolutely crazy. So, the other thing that they announced at the same time, which also is super super nice, is an update to anti-gravity. I've said for a while, anti-gravity is extremely good. It's free to use, right? And when Gemini 3 Flash first came out, especially, it was one of the most intelligent models I've ever used. Like even it it wasn't as good as, you know, Opus 4.7 or whatever, but like it was the the gap wasn't that large and the speed difference and kind of the multimodal abilities of the flash models and also just Gemini generally were what drew me to it. And the fact that it was completely free to use inside anti-gravity. So, the other thin [ASSISTANT] { "tldr": [ "Google IO dropped Gemini 3.5 Flash: 300 tokens/sec, near-Opus intelligence at flash speed, free inside Antigravity 2.0.", "Antigravity 2.0 adds new hooks, harness-agent-first design, subagents, and async management — full free IDE alternative to Codex/Claude Code for the next few weeks.", "Creator confirmed Gemini built an entire OS for $1K (mirroring the Opus 4.7 demo), positioning Flash as a serious agentic coding model.", "Install pattern: skills can be tagged directly inside Antigravity via terminal install + `@skill-name` tag — same mental model as Claude Code skills.", "Practical play: build throwaway/test projects in Antigravity on the free tier before Google changes pricing." ], "tools": [ { "name": "Google Antigravity 2.0", "url": "https://antigravity.google", "description": "Free agentic coding IDE with Gemini 3.5 Flash, subagents, async management, and hooks." }, { "name": "Gemini 3.5 Flash", "url": "https://deepmind.google/technologies/gemini/", "description": "New fast Google model, 300 tok/sec, free inside Antigravity." } ], "skill_candidates": [ { "slug": "antigravity-skill-installer", "description": "Pattern for installing and tagging Claude-Code-style skills inside Antigravity 2.0 via terminal + @skill tag." }, { "slug": "free-tier-agentic-build-router", "description": "Decision skill: route low-stakes builds to Antigravity+Gemini 3.5 Flash (free) vs Claude Code Opus (paid) based on task complexity." } ], "verdict": "worth-a-skim", "verdict_reason": "Hits agentic coding + LLM tooling domains and Antigravity 2.0's subagent/hooks/async features are worth a 5-min look, but it's mostly hype reaction with no deep workflow Mike doesn't already have via Claude Code." }

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: In The World of AI Video title: Gemini 3.5 Flash: Google's Most Powerful Model Ever! Beats Opus 4.7 & GPT 5.5? (Fully Tested) TRANSCRIPT (first 6000 chars): Google is finally back with the launch of Gemini 3.5 Flash officially, which launched today at the Google IO developer conference, and I've got some mixed thoughts about it. Gemini 3.5 Flash is Google's newest flash tier model, the fast and efficient lineup, but honestly, not so sure about how it is efficient anymore. But what's surprising is that this is a model that's positioned as Google's strongest agentic coding model yet, above the Gemini 3.1 Pro, which is interesting. This is a frontier level intelligence while still trying to keep the classic flash advantage with blazing fast speed, low latency, as well as cheaper realworld deployment. And to be fair, the model is great. The speed is insane. The coding is quite strong and the quality jump is very noticeable. But my issue is if the entire point of a flash model is to be cheap as well as fast, why did Google make it so that it is a token hungry model and quite expensive for a flash tier model? Cuz the Gemini 3.5 flash is costing you $1.50 50s per 1 million input tokens and $9 per 1 million output tokens. According to the intelligence index, it's over five times more expensive to run than the Gemini 3 flash and 75% approximately more costly than the Gemini 3.1 Pro on certain workloads. If you want the best AI tools, workflows, and drops before everyone else, join my free newsletter with the link in the description below, which is completely free. That said, technically this thing is very impressive. It is something that can plan and reason across massive code bases quite well, deploy sub agents in parallel over long horizon tasks, and it outperforms Gemini 3.1 Pro on benchmarks like Terminal Bench, GDP Evo, as well as MCP Atlas while still being remarkably close to many of these other proprietary giants like Claude Opus 4.7 as well as the GPT 5.5. But then according to artificial analysis, they basically state that overall wise in terms of its intelligence index, it is closer to Kim K 2.6, which is definitely much cheaper than Gemini 3.5 Flash. But then again, Gemini 3.5 Flash is extremely fast and better at reasoning. But then again, when I had looked at my own benchmark, which is something I'm going to be releasing soon, [ASSISTANT] {"tldr":["Google launched Gemini 3.5 Flash at I/O — positioned as their strongest agentic coding model, beating Gemini 3.1 Pro on Terminal Bench, GDP Evo, and MCP Atlas","Pricing is the catch: $1.50/M input, $9/M output — 5x more expensive than Gemini 3 Flash and ~75% pricier than 3.1 Pro on real workloads, breaking the 'cheap flash' promise","Token-hungry behavior: uses ~73M tokens vs GPT 5.5 medium's 22M on the same benchmark, while scoring slightly lower (55 vs 57)","Hallucination rate dropped from 91% to 61%, 1M context window, multimodal, Jan 2025 knowledge cutoff, free in Gemini app and AI Studio","Strong at front-end gen (SVG, Three.js, dashboards) and parallel sub-agent deployment for long-horizon tasks"],"tools":[{"name":"Gemini 3.5 Flash","url":"https://gemini.google.com","description":"Google's new agentic coding flash model, free in Gemini app, also via API"},{"name":"Artificial Analysis","url":"https://artificialanalysis.ai","description":"Cost-to-performance benchmark source cited for the token-hungry critique"}],"skill_candidates":[],"verdict":"worth-a-skim","verdict_reason":"Hits agentic coding domain and the pricing/token trap is useful intel for model routing decisions, but it's a benchmark review with no new tooling or workflow Mike can extract."}

[USER] You are an expert content analyst for Mike Merlino, an AI agency operator and builder. Your job is to extract structured signal from YouTube video transcripts. Mike's domains of interest (if a video hits 2+ of these → strong signal): [ "AI agents", "LLM tooling", "Claude Code", "MCP servers", "skills", "prompt engineering", "SEO automation", "GMB", "local SEO", "cold outreach", "SMS", "GoHighLevel", "Next.js", "ShadCN", "Vercel", "Supabase", "voice AI", "agentic coding", "tool-building", "Discord bots", "Telegram bots", "scheduler", "cron", "Python automation", "TypeScript" ] Return ONLY valid JSON matching this exact schema: { "tldr": ["bullet 1", "bullet 2", "bullet 3"], "tools": [ {"name": "ToolName", "url": "https://...", "description": "one line"} ], "skill_candidates": [ {"slug": "kebab-case-name", "description": "what skill this would capture"} ], "verdict": "dont-miss" | "worth-a-skim" | "skip", "verdict_reason": "one sentence explanation" } Verdict rules: - "dont-miss": new tool + active use-case for Mike OR novel pattern OR hits 2+ Mike domains OR introduces something Mike hasn't seen - "worth-a-skim": solid content, 1 Mike domain, no new tools - "skip": opinion/news/rehash with no actionable takeaway Be OPINIONATED. Do NOT default to neutral. Mike trusts your judgment. TL;DR bullets: 3-5, action-oriented, specific. No vague summaries. Tools: only real URLs you are confident about from transcript context. Omit if uncertain. Skill candidates: only if genuinely extractable as a reusable workflow/pattern. Channel: Eric W Tech Video title: Claude & Higgsfield MCP: 12 Ways to Automate ANYTHING! TRANSCRIPT (first 6000 chars): If you're a founder, running ads is honestly one of the most frustrating parts of building a product because most of the time you're just guessing. You make a few creatives, throw money to ad spend, wait for analytics, realize two of them completely failed, then repeat the process again next week. And for engineers especially, that whole workflow feels backwards because there's almost no automation in it. But MCP changes this quite a bit. Higgsfield just released their new MCP server, which basically gives AI agents direct access to cinematic media generation tools inside environments like Claude Code and Open Claude. So instead of AI just helping you brainstorm ad ideas, it can actually generate creatives, score them, [music] organize the outputs, and build a complete media pipeline directly inside your working directory. So in this video, we're going to build a fully autonomous AB ad engine for a fictional SaaS product called LogiTrace. The idea is pretty simple. We'll start with one boring screenshot of a dashboard UI. Then Claude Code connected to Higgsfield MCP is going to generate multiple cinematic video ads automatically, run them through an AI virality scoring pipeline, pick the mathematically strongest variant, and post the winning ad directly to Discord completely on its own. All right, so let's start by setting up the MCP connection between Claude Code and Higgsfield. The setup here is actually really simple because Higgsfield's MCP server is fully hosted, which means there are no API keys to manage. Authentication runs through OAuth directly in the browser. I'll be using Claude Code for this demo because I want the agent interacting directly with the file system and output folders while we work. So in the terminal, I'll just paste this in and run. Claude can see every Higgsfield tool exposed by the server. So now Claude isn't limited to just writing code or text anymore. It can actually generate production assets directly into the project itself. With the MCP connection working, let's build the actual pipeline. This is our starting point right here. It's literally just a flat screenshot of the LogiTrace dashboard. Functionally, it's fine, but from a marketing perspective, it's pretty boring. And honestly, this [ASSISTANT] { "tldr": [ "Higgsfield just shipped a fully-hosted MCP server (OAuth, no API keys) that exposes cinematic media generation tools to Claude Code and OpenClaw", "Demo builds an autonomous A/B ad engine: one dashboard screenshot → 3 cinematic image variants (Marketing Studio) → 3 video ads (C-DANCE 2.0) → AI virality scoring → auto-post winner to Discord", "Pattern worth stealing: generate multiple creative hypotheses programmatically, then score BEFORE spend instead of paying for bad experiments post-launch", "Claude orchestrates the pipeline (file discovery, MCP calls, output organization) without generating media itself — clean separation of agent vs generation layer", "Directly relevant to RankingReels / Creatify ad pipeline — Higgsfield MCP could replace or augment that flow with a virality-scored variant" ], "tools": [ {"name": "Higgsfield MCP Server", "url": "https://higgsfield.ai", "description": "Hosted MCP server exposing cinematic image/video generation (Marketing Studio, C-DANCE 2.0, Virality Predictor) to Claude Code via OAuth"} ], "skill_candidates": [ {"slug": "higgsfield-mcp-ad-engine", "description": "Autonomous A/B ad pipeline: screenshot → N cinematic image variants → N video ads → virality scoring → auto-publish winner. Wraps Higgsfield MCP tools with orchestration prompts."}, {"slug": "virality-pre-scoring", "description": "Score creative variants BEFORE ad spend using Higgsfield Virality Predictor. Replaces post-launch A/B testing with pre-launch mathematical selection."}, {"slug": "screenshot-to-cinematic-video", "description": "Convert flat product UI screenshots into cinematic SaaS commercial videos with parallax, lighting, and animated overlays via C-DANCE 2.0."} ], "verdict": "dont-miss", "verdict_reason": "New MCP server with hosted OAuth setup, directly hits Mike's domains (MCP servers, Claude Code, agentic coding, ad creative pipeline) and overlaps with RankingReels/Creatify workflow — the virality pre-scoring pattern alone is worth stealing." }