Descriptions:
Cole Medin demonstrates how he uses Claude Code, Obsidian, and a custom “Skills” architecture to build a personal second brain that handles research, content ideation, script drafting, slide generation, and external service integration. Rather than treating Claude Code as a coding-only tool, Medin has designed a modular system where the AI dynamically loads specialized instruction files depending on the task at hand.
The core architectural principle is progressive disclosure: each Skill is represented upfront only by a short description that Claude Code reads at session start. When the agent determines a Skill is relevant, it loads the full skill.md instruction file. Some Skills contain additional nested files for edge cases, creating a three-layer hierarchy that keeps context window usage low while allowing deep specialization. This same pattern is applied to MCP servers — Medin wraps Zapier’s MCP (which connects to Gmail, Google Calendar, Slack, and Asana) inside a Skill so the tool list doesn’t flood the context at startup.
Practical capabilities shown include generating Excalidraw diagrams, drafting YouTube scripts, creating on-brand PowerPoint presentations matched to Medin’s Dynamist branding, and pulling in external content from Obsidian notes and connected services. A starter template covering core Skills is linked in the video description and can be configured in five to ten minutes. Medin frames the system explicitly as an augmentation tool for ideation and organization rather than an automation engine for publishing AI-generated output.
📺 Source: Cole Medin · Published January 26, 2026
🏷️ Format: Workflow Case Study
