OpenSpec Output¶

Phase 4 produces the full specification as OpenSpec, a lightweight, capability-oriented format designed for AI coding agents.

What Gets Produced¶

Every Haytham artifact maps to a specific location in the OpenSpec output.

Haytham Artifact	OpenSpec Location
Capabilities (CAP-F-, CAP-NF-)	`specs/{domain}/spec.md` as SHALL statements
Acceptance criteria	Gherkin scenarios in `spec.md`
Architecture decisions (DEC-*)	`project.md`
Build/buy recommendations	`project.md`
System traits	`config.yaml`
Non-functional capabilities	`specs/cross-cutting/spec.md`

Directory Structure¶

openspec/
  config.yaml                  # Project metadata, system traits, appetite
  project.md                   # Tech stack, architecture decisions, build/buy
  specs/
    user-authentication/
      spec.md                  # SHALL statements + Gherkin scenarios
    leaderboard-management/
      spec.md
    cross-cutting/
      spec.md                  # Non-functional requirements

Using OpenSpec with Coding Agents¶

The OpenSpec output is designed to be dropped into a project root and consumed directly by coding agents.

Claude Code¶

Add the spec directory to your CLAUDE.md:

# Project Spec
See `openspec/` for the full specification.
Start with `openspec/project.md` for architecture context,
then implement specs in `openspec/specs/` order.

Other Agents¶

Any agent that reads markdown files from disk can consume the format. Point the agent at openspec/ and let it discover the structure.

Links¶

OpenSpec repository