5 GitHub Repos: Kill AI Slop, Go Viral, Make Money
Greg Isenberg · 2026-09-02
💡 Quick Take
1. Use Peter Yang's No AI Slop Skill via NPX to strip synthetic patterns from AI-generated writing while preserving your authentic human voice.
2. Use trycompai CRM as a local workspace agent (requires Bun and Docker) to maintain your relationship graph and automatically follow up on warm leads.
3. Use video use with coding agents like Claude Code or Codex to automate repetitive video editing workflows, cut dead space, and generate content formats.
4. Use Nvidia's Skill Specter via UV to scan third-party AI agent skills and MCP servers for security vulnerabilities, prompt injection, and hidden risks.
5. Use phone harness with Mac iPhone mirroring or Android ADB to let coding agents control real mobile devices for automated testing and workflows.
📊 Detailed Explanation
The first featured repository is Peter Yang's No AI Slop Skill, designed to combat the pervasive issue of formulaic, overly polished, and symmetrical text generated by AI models like ChatGPT and Claude. The speaker notes that generic AI writing carries a specific smell—using predictable tropes like concluding phrases or filler words like quietly—which erodes trust even if grammar and syntax are technically correct. Rather than stripping personality away like standard cleaning tools, this skill hunts for robotic phrasing patterns in rough human drafts to make communication more believable. The speaker advises installing it via the command line with NPX skills add and running it as a post-processing step after you have already written your raw ideas and outlines.
The second tool is trycompai CRM, an open-source customer relationship management platform built specifically for AI agents rather than human data-entry filing cabinets. Traditional CRMs frequently fail because users stop updating data, leading to stale notes and lost revenue sitting dormant in past email threads and waitlists. This open-source alternative uses an agent to actively research, enrich company info, schedule follow-ups, and maintain relationship graphs automatically. Setup requires technical prerequisites including Bun and Docker, cloning the repository, configuring environment variables for Google or Microsoft authentication, and running database deployment scripts to host it locally on localhost:3000.
The third tool, video use by browser use, enables creators and founders to edit raw video footage using coding agents like Claude Code or Codex. Content creation is a major bottleneck for businesses, yet editing requires repetitive execution of standard cuts, subtitle generation, and pacing adjustments. By pairing raw footage folders with an agent, users can automate filler word removal, color grading, and video rendering based on explicit transcripts. The speaker recommends starting small rather than trying to automate an entire channel on day one—such as turning a founder's Loom video into a 60-second clip or extracting podcast highlights before scaling it into a repeatable system.
The fourth repository is Skill Specter, released by Nvidia, which addresses the growing security risks of installing random AI agent skills, plugins, and Model Context Protocol (MCP) servers directly from GitHub. Because these downloaded capabilities can execute code, touch files, and connect to external services, running unverified scripts introduces supply chain risks, prompt injection vulnerabilities, and data exfiltration threats. Skill Specter allows users to scan local directories or GitHub links for malicious patterns before granting tools to an AI agent. Installation can be done quickly via UV tool install, and users can opt for a --no-lam flag to perform static scans without transmitting sensitive file contents to external language model providers.
The fifth tool is phone harness, which allows AI agents to directly control physical iPhone and Android devices to execute mobile workflows and user interface testing. While most agent tools focus on desktop browsers and terminals, numerous critical applications lack APIs and exist exclusively behind mobile interfaces. For iPhones, the repository leverages native macOS iPhone mirroring, while Android integration relies on standard Android Debug Bridge (ADB) connections over USB or wireless without requiring jailbreaking or specialized mobile applications. The speaker points out that while the tool is still early and faces challenges with Face ID or camera interactions, it opens up immense practical utility for automated mobile QA testing, onboarding checks, and exploring new mobile-first automation agency ideas.
🎯 Tech Expert Opinion
The curated selection of GitHub repositories in this episode captures the exact shift happening in modern software development, moving away from static web applications toward autonomous, agent-driven workflows. By highlighting open-source projects that address concrete friction points—such as unauthentic AI writing copy, neglected CRM pipelines, video editing bottlenecks, supply chain security for agent tools, and mobile device automation—the speaker provides a practical roadmap for technical and non-technical early adopters alike. Crucially, the recommendations emphasize actionable execution: starting with narrow, well-defined workflows rather than attempting full automation straight out of the gate.
A notable strength of the presentation is its realistic framing of technical barriers and early-stage limitations. Tools like trycompai CRM and phone harness require local development environments, command-line operations, environment variable management, and specific operating system dependencies such as Docker, Bun, or macOS Sequoia. The speaker does not sugarcoat these requirements as one-click consumer apps, correctly identifying them as powerful, product-shaped developer repositories. This transparency prevents novice users from blindly pushing unstable setups into production environments while encouraging builders to experiment locally and validate utility.
Ultimately, these repositories serve as an invaluable early-warning system for upcoming commercial SaaS trends and micro-startup opportunities. Security tools like Nvidia's Skill Specter underscore a vital industry pivot toward agent safety and sandboxing as users increasingly grant autonomous code execution permissions to external repositories. Whether utilized for personal productivity gains, content creation pipelines, or agency service offerings, the suggested tools demonstrate that mastering GitHub discovery in the agentic era yields an undeniable unfair advantage for builders willing to get their hands dirty.
Kanal: Greg Isenberg