Summary
Sharbel A. walks through ten modular “Hermes agent skills” — drop-in capability files and MCP servers — organized into three groups: skills that cut token costs, skills that add new abilities, and skills that turn a single agent into a coordinated team.
On the cost side, Defuddle strips web pages to clean markdown before the agent reads them, eliminating navigation bars and ads that inflate context length. Caveman applies ultra-compressed output formatting the creator benchmarks at roughly a 65% reduction in output tokens. A third tool, Codebase Memory, indexes entire repositories into a graph so the agent queries a map instead of reading files repeatedly — Sharbel reports 95% fewer tokens in his own test against the repo’s claimed 99%. For new capabilities, Agent Reach unlocks X, Reddit, GitHub, and other social platforms without API fees and includes automatic failover routing. The Marketing Skills pack bundles 49 frameworks covering SEO, paid ads, copywriting, and buyer psychology into a single install. Compozio connects Gmail, Google Sheets, Slack, Notion, HubSpot, Salesforce, and over a thousand other tools without manual OAuth setup.
The video also covers Humanizer, which rewrites AI-generated text to remove detectable patterns using a Wikipedia-sourced list of AI writing tells, and Open Montage, described as the first open-source agentic video production system. Sharbel ends with a clear recommendation on which three skills to install first, making this a practical starting point for anyone building cost-efficient, capable AI agents.
📺 Source: Sharbel A. · Published August 16, 2026
🏷️ Format: Tutorial Demo
