S4: A Framework for AI‑Driven Software Development

S4 is a CLI tool that enables stateless AI-human collaboration through semi-structured software specifications. By maintaining all project context in a single s4.yaml file, it allows each development session to be completely self-contained - no chat history required, no context decay, just specification-driven iteration.

S4 is a command-line tool designed specifically for AI-driven software development. It solves the "context rot" problem that plagues AI coding assistants by providing complete project context on demand.

When working with AI coding assistants, you've probably experienced:

• Context decay: AI forgets earlier decisions as conversations get longer
• Lost continuity: Starting a new chat means losing all project context
• Token waste: Copying entire codebases into context windows
• Inconsistent results: AI makes different decisions in different sessions

S4 maintains your project's complete context in a structured s4.yaml specification file. When you run s4 status, it analyzes both your specification and codebase to provide:

• Complete project overview (mission, vision, business objectives)
• Feature completion status (which acceptance tests pass/fail)
• Actionable next steps (exactly what needs to be done)
• Validation issues (spec inconsistencies that need fixing)
• Tool execution results (linting, testing, etc.)

npm install -g s4

1. Create a specification (s4.yaml):

title: My Awesome Project
mission: Build a user-friendly task management app
vision: Help teams collaborate more effectively

businessObjectives:
  - id: BO-0001
    description: Enable users to create and manage tasks

features:
  - id: FE-0001
    title: Task Creation
    description: Users can create new tasks with title and description
    covers: [BO-0001]

acceptanceTests:
  - id: AT-0001
    covers: FE-0001
    given: a user wants to create a task
    when: they click the "New Task" button
    then: a task creation form appears

2. Check project status:

s4 status

3. Follow the guidance - S4 will tell you exactly what to do next!

The main command that provides complete project context and next actions.

s4 status

Validates your specification for internal consistency.

s4 validate

Provides guidance on writing specifications.

s4 guide  # Show brief guidance
s4 guide feature  # Show feature writing guide
s4 guide acceptanceTest  # Show acceptance test guide

Shows detailed information about features or acceptance tests.

s4 info FE-0001  # Show feature details
s4 info AT-0001  # Show acceptance test details

Your s4.yaml file contains:

• Project metadata: title, mission, vision
• Business objectives: High-level goals (BO-0001, BO-0002, etc.)
• Features: Concrete functionality (FE-0001, FE-0002, etc.)
• Acceptance tests: Testable conditions (AT-0001, AT-0002, etc.)
• Concepts: Reusable definitions referenced with [[concept]]
• Tools: Custom commands to run (linting, testing, etc.)
• Connectors: How to find and run your tests

1. Traceability: Every business objective must be covered by features, every feature by acceptance tests
2. Self-contained: Each acceptance test should be independently testable
3. AI-friendly: Specifications are written for both humans and AI agents
4. Validation: S4 ensures your specification is internally consistent

Modern AI coding assistants treat conversation context like critical state that must be preserved across interactions. This creates cascading problems:

• Context decay: AI performance degrades as conversations get longer
• Lost continuity: New chat sessions lose all project context
• Token waste: Copying entire codebases into context windows
• Inconsistent results: AI makes different decisions in different sessions

S4 recognizes that well-crafted specifications are the primary artifact in AI-assisted development. The code becomes a derived artifact—important for execution, but secondary for capturing intent.

The framework enables truly stateless AI interactions by maintaining all intent, context, and requirements in a persistent specification. The CLI analyzes this specification alongside your code to generate complete, self-contained context on demand.

Every development session should be independent. Previous chat history becomes irrelevant when s4 status can provide complete context every time. This isn't just about convenience—it's about reliability and reproducibility.

S4 maintains a clear separation of concerns:

• CLI: Handles deterministic analysis of structure, traceability, and coverage
• AI agents: Interpret intent and execute recommendations
• Humans: Make strategic decisions and set direction

1. Human updates s4.yaml (possibly with AI assistance)
2. Open fresh AI chat and ask AI to run s4 status
3. AI gets complete context and completes the recommended task
4. Human closes chat without losing any context or progress
5. Repeat with any AI at any time

See the examples/ directory for real-world specification examples.

MIT