TL;DR: SDD puts the specification at the center: well-written specifications can be used to generate code via AI agents. Fewer misunderstandings, faster iteration.

Definition of the concept​

Imagine the specification is the recipe and the code is the final dish. In Specification Driven Development (SDD), the recipe is not a poor note in the margin; it's the manual that the chefs (including robot chefs—LLMs) use to cook the application.

SDD emerges as a philosophy in response to LLM-based agents and vibe coding: instead of writing code first and then documentation, well-written and executable documentation becomes the "source of truth" and guides code generation.

Quick example: you write a specification: "When button X is pressed, send notification Y to the user with conditions Z". An agent can transform that specification into endpoints, tests, and minimum viable deployment. Simple, repeatable, and traceable.

Source of truth​

• Agile / Waterfall: Code rules. Specs often become outdated notes.
• SDD: The specification rules. Code is a temporal expression of that specification: if something is off, the spec is the source to correct.

In SDD, the specification functions as a contract: when the spec and code differ, the spec is the truth until we decide otherwise (and document it).

Change management (by analogy)​

Think of changes as updating a recipe, not re-making the dish by hand every time:

• Before: a change in requirements causes chaos: code editing, PRs, hotfixes.
• With SDD: change the spec, regenerate implementations and tests. Experiment what-if at low cost.

Why now? (trends pushing SDD)​

1. More capable AI: natural language specs generate functional code reliably.
2. More complex software: maintaining consistency manually is increasingly expensive.
3. Accelerated pace of change: pivots are the norm and you need to iterate quickly and safely.

This also enables an interesting game: generate multiple implementations from the same spec and compare advantages (performance vs. cost vs. maintainability).

Fundamental principles (one sentence each)​

1. Specifications as lingua franca — the spec is the central artifact.
2. Executable specifications — precise, complete, and unambiguous.
3. Continuous refinement — validation and ambiguity detection all the time.
4. Research-driven context — agents collect context and technical options.
5. Bidirectional feedback — production feeds back into the specification.
6. Branching for exploration — generate alternatives to optimize different objectives.

Small exercise (try it!)​

Take a user story and turn it into a minimal spec (3–5 lines) that's clear and unambiguous. Then ask yourself:

• Can an agent generate endpoints and tests from this?
• What ambiguities remain?

If you answered yes and there's little ambiguity, you've taken a big step toward SDD.

Final reflection​

SDD doesn't come to replace the developer: it comes to turn them into a specification architect, orchestra conductor, and curious critic. It's a powerful way to accelerate iterations, improve traceability, and minimize technical debt when properly implemented.

Sources and references​

• SDD definition provided by GitHub Spec Kit: https://github.com/github/spec-kit/blob/main/spec-driven.md

This post explains how to use the npx skills command (the skills manager) and how to integrate it into your workflow in Visual Studio Code.

What is npx skills?​

npx skills is the CLI for the skills ecosystem: it helps you search, install, and update packages called skills that extend the agent's capabilities.

Key commands:

• Search: npx skills find [query]
• Install: npx skills add <owner/repo@skill>
• Update: npx skills update

Quick usage from VS Code​

1. Open the integrated terminal (View → Terminal or `Ctrl+``).
2. Search for relevant skills:

# search for skills by keywords
npx skills find react performance

3. Review the output and, if interested, install one:

# install a skill (global installation is optional, you can use npx directly)
npx skills add vercel-labs/agent-skills@vercel-react-best-practices

4. Optional: install globally and avoid typing npx:

npm i -g @skills/cli
skills find "react testing"

Practical tip​

If you want to work with skills frequently from VS Code, keep the terminal pinned and use searches with specific words (e.g., react testing, docs, deploy).

Example flow​

• Search: npx skills find changelog
• Install: npx skills add composio/changelog-skill
• Execute the task the skill provides: depends on the installed skill (check the documentation at skills.sh).

Resources​

• Website: https://skills.sh/ 🔗
• Documentation and examples (use npx skills find to discover specific skills)

If you want, I can install a sample skill here and show step-by-step how to use it. Which category interests you (tests, docs, deploy, productivity)?