rickmanelius/skills: Compiling a Business Book into an AI Mentor
How structured Markdown and the open Agent Skills standard turned passive startup advice into an executable, stateful virtual co-founder.
- The repository transforms static business advice into an executable state machine using the open Agent Skills standard.
- It engineers cognitive friction by forcing founders through rigorous conversational tollgates before generating any solutions.
- By separating context data from execution logic, the project treats natural language prompting with the structural rigor of traditional software architecture.
The Architecture of Friction
Most AI tools are designed to reduce friction. They take a vague prompt and immediately generate a polished artifact. This repository takes the opposite approach. It relies on a five-phase Wizard pattern to actively stop the AI from hallucinating a final product too early.
Skills like sys-paint-done and sys-assumptions-audit act as cognitive tollgates. The AI is explicitly instructed to ask clarifying questions and evaluate assumptions before it is allowed to generate the final business framework. This prevents the common failure mode of AI sycophancy, where the model simply gives the user what they want instead of what they need.
The Open Standard Under the Hood
This workflow is powered by the Agent Skills standard originally published by Anthropic. A skill is essentially a directory containing a SKILL.md file. This file uses YAML frontmatter for metadata and Markdown for execution instructions.
The standard relies on progressive disclosure to save context window space. Tools like Claude Code only read the tiny YAML frontmatter at startup. The heavy Markdown instructions are loaded into memory only when the user explicitly triggers the skill.
a skill is not a configuration file. It is not a prompt template. It is a procedure - a set of instructions that tells an agent how to accomplish a specific task the way your team does it.
Separation of Concerns in Natural Language
A look at the repository structure reveals a mature approach to prompt engineering. The author separates the raw business knowledge base from the execution logic. The context/ directory holds chapter breakdowns and master rankings. The plugins/ directory contains the actual state machines.
This mimics the Model-View-Controller architecture of traditional software. The data layer is decoupled from the runtime environment. Prompt engineering is no longer a monolithic paragraph pasted into a web interface. It is a modular, version-controlled file tree.
Conversational State vs. The GUI Wizard
Traditional software handles complex data entry via multi-step graphical wizards. These forms rely on hardcoded validation rules and rigid linear paths. The approach in this repository replaces the form with a conversational thread.
| Feature | Traditional GUI Wizard | Agent Skill Wizard |
|---|---|---|
| State Management | Database flags and session storage | LLM context window phases |
| Friction | Validation errors after submission | Real-time clarifying dialogue |
| Flexibility | Rigid linear paths | Dynamic branching based on comprehension |
By codifying a business book into an interactive CLI tool, the project demonstrates a new paradigm for both software distribution and startup mentorship. The code is English, the compiler is an LLM, and the output is a more focused founder.