The open specification standard for AI-driven development
SpecScore is an open specification format that makes requirements machine-readable without making them human-unreadable. Markdown and YAML — version-controlled, portable, no vendor lock-in. A linter catches ambiguity before your agents do.
The Problem
AI agents work from specifications. Most specifications are ambiguous, scattered across Jira tickets and Notion docs, and impossible to validate automatically. When agents guess what you meant, you get rework. When specs live in chat logs, you get context loss.
The Format
- Feature Structure and metadata for a single feature specification
- Acceptance Criteria Machine-verifiable conditions that define feature completeness
- Plan Structured task breakdowns that bridge specs to execution
- Task Discrete units of work with status lifecycle and dependencies
- Source References Annotations that link source code back to the governing spec
- Project Definition Project-level configuration, structure, and metadata
The Tooling
Write specs. Validate them. Ship them.
Find Your Path In
- Developers Clear specs, source references, structured features
- Product Owners Standardized format for vision, goals, acceptance criteria
- QAs Testable acceptance criteria and full feature context
- Business Analysts Formalize analysis into actionable, traceable specs
- Project Managers Structured scope, task breakdowns, estimation visibility
- Architects Technical decisions and system design documented consistently
Ecosystem
SpecScore defines what gets built. When you're ready to test specs automatically or orchestrate agents across them, the ecosystem is there. But SpecScore works beautifully on its own.
- Rehearse Test your specifications automatically
- Synchestra Spec-driven orchestration for AI-assisted development