The Agentic Socratic: A Skill for Intent-First Development
Most software projects fail because we start coding before we have finished thinking. I wanted to solve this by building a tool that forces the "Why" and the "How" to be fully articulated before the first line of code hits the disk. Last week, I built and shipped a portable, version-controlled Socratic agent skill in a single session. It proves that professional-grade agentic workflows can be shipped as lean, executable specifications rather than complex, opaque dependencies.
Quick notes on process
I started with a clear architectural constraint: stop rushing to boilerplate or npm install before the business logic is modeled. I needed a skill—a portable unit of logic—that acts as a Socratic interviewer, forcing me to map my domain logic before I commit to an implementation.
The Recursive Questionnaire: I built a three-round interactive loop where the agent iterates on its own design. I didn't just tell it what to do; I forced it to define its own persona and output schema, refining the protocol until the logic was watertight.
The Scaffold: I structured the logic into four linked Markdown files: SKILL.md (the core logic), INTERVIEW.md (the protocol), SYNTHESIS.md (the reasoning engine), and OUTPUT.md (the delivery). This keeps the design concerns decoupled.
The Deployment: I bypassed the browser, pushing the files directly from the IDE to GitHub via the GitHub API. This keeps the repository in a constant state of sync with my local environment, managed autonomously via my preferred choice of harness.
What the Skill Does
This skill forces clarity by enforcing a strict design protocol. It refuses to assume your architecture. It extracts intent and synthesizes it into a mission document, an agent profile, and a logic scaffold—the exact input an LLM needs to handle the implementation phase autonomously. Whether you’re working in Cursor, Windsurf, or a custom CLI, the output is ready for immediate handoff to an execution agent.
How to use it:
The skill is hosted on GitHub. Clone the repo to your local workflow:git clone https://github.com/Kimotep/skills.git
To execute, load socratic-agentic-workflow/SKILL.md into your agent’s context window or paste it as a system prompt. The logic is platform-agnostic, designed to transform any chat interface into a rigorous design session.
The Dogma of Open Standards
This isn't about platform-specific locking. The skill format I’ve adopted is an open standard. By keeping the logic portable and versioned, I ensure that my infrastructure remains resilient, regardless of which LLM provider or IDE plugin dominates the market next month. This is the "API of Self" in practice: my thought process, codified and machine-executable.
What's Next
I’ve published this under an MIT license. It is a work in progress, intentionally lean, and ready for integration. You can find the source here: github.com/Kimotep/skills.
My objective is to build a library of high-signal, agentic workflows that minimize the gap between "intent" and "scaffolded system." If you’re building in this space, send a PR. Let’s evolve the standard together.