PBHQ Original Engine

VisionSpec

The spec engine — multi-domain specification orchestration for humans and AI agents. VisionSpec ships an embedded catalog of workflow methodologies — AWS Working Backwards, Big Tech, PBHQ Lite, and any workflow you define — and executes them directly, turning organizational intent into specs an agent can build against.

Where it sits. VisionSpec is both the contract and the engine — workflow definitions (types, templates, rubrics) and the code that runs them, in one repo. VisionStudio is the desktop studio that imports it. One engine, one studio.

Three Surfaces

The same engine, exposed three ways — so it meets you wherever you work: in code, in a terminal, or inside your coding agent.

Importable SDK

Go library

Go packages that load workflows, synthesize specs, run LLM-as-a-Judge evaluation, and report status. This is the layer VisionStudio imports directly — no shelling out to a binary.

import "github.com/ProductBuildersHQ/visionspec/pkg/..."

CLI

visionspec

A command-line engine for people and terminal-native agents: scaffold a project on a chosen methodology, validate structure, synthesize documents, and emit a readiness report.

visionspec init my-product --workflow=aws-working-backwards/product

MCP Server

cmd/mcp-server

A Model Context Protocol server that exposes the same engine to any coding agent — Claude Code, Cursor, Copilot — so an agent can author, synthesize, and evaluate specs inside the workflow it already runs.

go run github.com/ProductBuildersHQ/visionspec/cmd/mcp-server

Two Ways to Use It

Directly by an AI agent

Point a coding agent at the MCP server (or the CLI) and it authors, synthesizes, and evaluates specs inside the workflow — no new app to adopt, just a tool your agent already knows how to call.

Imported into VisionStudio

VisionStudio imports the VisionSpec SDK directly to give the same engine a desktop studio — authoring, evaluation dashboards, and execution tracking over one project.

What the Engine Does

  • Working Backwards flow — start from a vision (Press Release), derive requirements (PRD)
  • Dual-methodology system — separate requirements (WHAT) from implementation (HOW)
  • LLM synthesis — generate Press, FAQ, PRD, TRD, TPD, IRD from source specs
  • LLM-as-a-Judge evaluation — per-domain judges with numeric scoring
  • Reconciliation — conflict detection and tradeoff resolution into a single execution spec
  • Target adapters — export to Spec-Kit, AWS AI-DLC, OpenSpec, and more
View on GitHub Browse the SDD Catalog