AI can write code. But production-grade software needs more than code generation.
Most AI-assisted development suffers from three critical gaps:
| Gap | The Problem | Our Solution |
|---|---|---|
| Ambiguity | Vague prompts lead to implementations that miss the target | Structured specifications as single source of truth |
| Traceability | Generated code is disconnected from business requirements | End-to-end linking from PRD to architecture to code |
| Validation | Code ships without proper testing or verification | Automated test generation and requirement verification |
A multi-agent orchestration system for AI-native software development. Delivers multiple SDLC flows as markdown-based agent systems - from Simple (quick tasks) to Fire (rapid features) to full AI-DLC (enterprise workflows). AI drives the conversation, humans validate.
- Multi-Agent Orchestration - Specialized agents working in concert for each flow
- Pluggable SDLC Flows - Choose your workflow:
- Simple - Spec generation for prototypes, MVPs, and team handoffs
- FIRE - Rapid feature development with adaptive checkpoints for brownfield/monorepo projects
- AI-DLC - Complete lifecycle with DDD/TDD/BDD for complex domains and regulated environments
- Rapid Iterations - Development cycles measured in hours, not weeks ("Bolts")
- Human-in-the-Loop - Validation at every stage ensures quality
- Universal Compatibility - Works with 10+ AI tools
Supported Tools
# Get started in seconds
npx specsmd@latest install
"We need automobiles, not faster horse chariots."
Traditional software development methodologies were designed when humans wrote every line of code. That world no longer exists. AI has fundamentally changed how software is created, yet most teams still retrofit AI into processes designed for a different era.
┌─────────────────────────────────────────────────────────────────────────────┐
│ │
│ 2020-2023 2023-2025 2025+ │
│ ────────── ────────── ──── │
│ AI-ASSISTED AI-DRIVEN AGENTIC │
│ │
│ Autocomplete AI generates AI autonomously │
│ Suggestions Humans approve executes workflows │
│ Tab-completion Code review focus Human validates │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
| What Worked Before | Why It Fails Now |
|---|---|
| Two-week sprints | Artificially slow when AI can generate in hours |
| Story point estimation | Meaningless when code generation cost collapses |
| Manual task breakdown | AI can decompose work more consistently |
| Sprint-bound context | Context lost; AI needs persistent memory |
Learn More
-B31B1B?style=flat-square&logo=arxiv&logoColor=white){kind=icon}
Open-source utilities for the AI-native developer workflow.
|
Multi-agent orchestration CLI Install and run specs.md flows directly from the command line. Supports Simple, FIRE, and AI-DLC workflows. npx specsmd@latest install
|
specs.md sidebar for VS Code Manage your specs, intents, and work items directly from VS Code with a dedicated sidebar. code --install-extension fabriqaai.specsmd
|
|
Obsidian-like editing in VS Code Three editing modes: Live Preview, Source, and Reading. Mermaid diagrams & GFM support. code --install-extension fabriqaai.fabriqa-markdown-editor
|
Search your Claude conversations Transforms JSONL chat logs into a searchable web interface. Full-text search, syntax highlighting. brew tap fabriqaai/tap && brew install claude-code-logs
|
"AI agents thrive when we give them structure, constraints, and intent."
We believe the future of software development is spec-driven. Writing clear specifications before generating code transforms AI from a code completion tool into an intelligent development partner.
| Traditional AI Prompting | Spec-Driven Development |
|---|---|
| Vague, one-shot prompts | Structured requirements |
| Multiple iterations to get it right | Right the first time |
| Context lost between sessions | Full traceability |
| Hope it works | Validated output |
A detailed spec acts as the single source of truth, enabling AI to make intelligent decisions rather than guessing at requirements. This is the foundation of everything we build.
Read about our journey building AI-native development tools:
- Announcing claude-code-logs - A searchable archive for your Claude Code conversations
- llm-cli: Simple Aliases - Streamlining your LLM workflow
We build in the open. All our projects are MIT licensed and contributions are welcome.
| Use | Try specs.md for your next project |
| Join | Chat with us on Discord |
| Follow | @specsmd and @hancengiz on X |
