SpecPrompt Documentation

Prompts are transient. Specs persist.

Part of the [&] Protocol stack · Ecosystem overview · Three-protocol stack · Stack status

Welcome to the documentation hub for SpecPrompt — an open standard and toolchain for spec-driven AI development. SpecPrompt defines a Markdown-based specification format (SPEC.md) that is both human-readable and machine-parseable, serving as the source of truth for agent behavior.

SpecPrompt is the standards layer of the [&] Protocol ecosystem — the format that bridges agent definition with execution, testing, and deployment.

Why SpecPrompt?

Spec-Driven Development (SDD) has emerged as the industry consensus for AI-assisted development in 2026. GitHub Spec Kit, AWS Kiro, Cursor rules, and Next Moca ADL all use different specification formats. Teams switching tools lose context.

SpecPrompt defines the interoperable standard — a Markdown-native format that is:

Markdown-native — .md files, no proprietary format
Human-first — readable by any team member, parseable by any agent
Versioned — lives in git. Branching, diffing, merging.
Testable — acceptance criteria are machine-executable
Composable — specs reference other specs
Tool-agnostic — works with any SDD-compatible tool

Documentation Map

Homepages

Root Docs

SpecPrompt Docs

README

The SPEC.md Format

Every spec follows a structured Markdown format with YAML frontmatter:

---
name: agent-name
version: 1.0.0
runtime: opensentience | any
author: team-or-person
created: 2026-01-15
updated: 2026-02-22
tags: [category, domain]
dependencies:
  - graphonomous
---

## Purpose
What this agent does and why.

## Capabilities
- capability:scope (description)

## Constraints
- Hard and soft behavioral boundaries

## Acceptance Tests
- Given [precondition] -> [expected behavior]

Required sections: Purpose, Capabilities, Constraints, Acceptance Tests.

Toolchain

CLI

specprompt validate SPEC.md      # Validate against format standard
specprompt lint SPEC.md          # Check for best practices
specprompt test-gen SPEC.md      # Generate test scaffolding
specprompt diff SPEC.md@v1 SPEC.md@v2  # Diff two versions
specprompt publish SPEC.md       # Publish to FleetPrompt registry

MCP Tools

Tool	Description
`spec_validate`	Validate against format standard
`spec_lint`	Check for best practices
`spec_generate`	Generate from natural language
`spec_test_gen`	Generate test cases from acceptance criteria
`spec_diff`	Compare two spec versions
`spec_search`	Search FleetPrompt registry

Reference Parsers

Available in Elixir and TypeScript. Formal PEG grammar included in the specification for building conformant parsers.

Ecosystem Integration

Product	How It Uses SpecPrompt
Agentelic	Specs are the primary input to the build pipeline
OpenSentience	Agent manifests reference SPEC.md for permission derivation
Graphonomous	Learning happens within spec-defined boundaries
FleetPrompt	Specs are published as discoverable templates
Delegatic	Multi-agent orchestration specs define agent roles

Interoperability

AGENTS.md — complementary (codebase conventions vs. agent behavior specs)
Next Moca ADL — bidirectional import/export
ampersand.json — bidirectional mapping (capabilities, constraints, governance)

Project Links

Spec: Product Specification
User stories: SpecPrompt User Stories
Build prompt: SpecPrompt Implementation Build Prompt
[&] Protocol ecosystem: AmpersandBoxDesign/

[&] Ampersand Box Design — specprompt.com