AI-native specification infrastructure for modern software teams.
SpecAgnt is not another AI agent framework.
It is a specification and validation layer between ideas, documentation, AI coding agents, and production codebases.
SpecAgnt generates structured, AI-readable project specifications and continuously validates that documentation, architecture, APIs, security, and implementation stay aligned over time.
v2.0 Update: SpecAgnt is now an Agent Lifecycle Framework. It goes beyond documentation to define the identity, personality, and evolution of the AI agents that live within your project.
Most AI tooling focuses on:
- agent orchestration
- code generation
- prompt workflows
- autonomous execution
SpecAgnt focuses on:
Specification consistency, AI interoperability, and codebase governance.
SpecAgnt validates consistency across generated artifacts.
Validation includes:
- feature ↔ API mapping
- architecture consistency
- security coverage
- deployment dependency validation
- orphaned requirements detection
- missing service relationships
- undocumented endpoints
- inconsistent auth strategies
The validation engine treats documentation as executable infrastructure.
SpecAgnt is not write-only documentation.
It can inspect existing codebases and detect specification drift.
Examples:
- API route exists but missing in
api-spec.md - Authentication changed but
security-compliance.mdis outdated - Database schema diverges from
data-model.md - Services exist in architecture but not deployment topology
- Features implemented without PRD coverage
This creates a continuous synchronization layer between:
Docs ↔ Architecture ↔ Runtime ↔ Codebase ↔ AI Agents
Every generated project exposes:
/.well-known/ai-manifest
The manifest is:
- MCP-compatible
- agent-readable
- IDE-readable
- CI-readable
- automation-friendly
SpecAgnt treats the manifest as a machine-readable contract between applications and AI systems.
SpecAgnt follows:
Vision → Research → Architecture → Security → Operations → Validation
Each layer validates upstream decisions before downstream generation continues.
SpecAgnt/
├── engine.md
├── validation/
│ ├── validation-engine.md
│ ├── drift-detection.md
│ └── consistency-rules.md
├── schemas/
│ └── ai-manifest.schema.json
├── base/
├── knowledge/
├── scripts/
└── examples/
The validation engine performs:
| Validation | Purpose |
|---|---|
| Feature Coverage | Ensure all features map to APIs |
| Security Coverage | Ensure exposed APIs are protected |
| Dependency Integrity | Detect missing services or infra |
| Architecture Consistency | Validate topology alignment |
| Drift Detection | Compare codebase against specs |
| Manifest Integrity | Validate AI manifest correctness |
SpecAgnt continuously compares:
| Source | Compared Against |
|---|---|
| Routes | api-spec.md |
| Database schemas | data-model.md |
| Auth middleware | security docs |
| Infra configs | deployment docs |
| Services | architecture diagrams |
The manifest is designed to support:
- MCP tooling
- AI IDEs
- autonomous agents
- CI/CD systems
- runtime discovery
- tool invocation contracts
- AI capability negotiation
{
"manifest": "ai-manifest",
"version": "2.0",
"mcp_compatible": true,
"agent_readable": true,
"capabilities": {
"features": [],
"tools": [],
"resources": []
}
}- CLI validator
- spec graph engine
- manifest linter
- GitHub Actions integration
- IDE extension support
- schema auto-generation
- live architecture diffing
- runtime topology inspection
- autonomous remediation suggestions
- multi-agent coordination protocol
- AI-native governance workflows
SpecAgnt is best understood as:
“Executable documentation, validation infrastructure, and Agent Lifecycle Framework for AI-assisted software delivery.”
Not another coding agent.
MIT