Overview

The Forge Skills Starter is an AI Resources Lifecycle Platform — a convention-over-configuration template for building, testing, and delivering AI agent skills, rules, and workflows.

What is an AI Resource?

An AI resource is any artifact that shapes how an AI assistant behaves — a skill, a rule, a workflow, or an agent definition. Resources are versioned, tested, and delivered to developer environments through structured conventions.

The platform treats resources as first-class engineering artifacts with a full lifecycle: design → develop → test → release → deliver.

How it works

One orchestrator, three phases:

1. Design — /design-resource creates design docs and architecture diagrams
2. Develop — /develop-resource scaffolds and implements from the design
3. Test — /test-resource runs type-aware testing against real project specimens

/create-resource orchestrates all three end-to-end.

Convention over configuration

Put a file in the right place — the platform discovers it. No config, no registration.

What	Convention	Discovered by
Skill	skills/<name>/SKILL.md	/create-resource, /test-resource, Forge
Skill docs	skills/<name>/README.md	Developers browsing the repo
Design	docs/<name>/design.md	/design-resource, /develop-resource
Rule	rules/<name>.md	/create-resource, Forge
Workflow	workflows/<name>.md	/create-resource, Forge
Specimen	orphan branch specimen/<lang>/<fw>	/test-resource

Platform structure

Repository structure

forge-skills-starter/
├── skills/                   # Shared skills (delivered via Forge)
│   ├── drawio/
│   ├── detect-stack/
│   ├── review-pr/
│   └── document/
├── rules/                    # Shared rules (delivered via Forge)
├── workflows/                # Shared workflows (delivered via Forge)
├── docs/                     # Platform docs (NOT delivered)
├── scripts/                  # Test automation (NOT delivered)
└── .claude/skills/           # Meta-skills (NOT delivered)
    ├── create-resource/
    ├── design-resource/
    ├── develop-resource/
    └── test-resource/

Shared (skills/, rules/, workflows/) → delivered to consumer repos via Forge.

Platform (everything else) → stays in this repo, powers development and testing.

Multi-team / Monorepo setup

The main branch is a single-team setup. For organizations with multiple teams, the example/monorepo branch demonstrates team-scoped resources:

shared/       # Cross-team resources (every team gets these)
frontend/     # Frontend team — /review-frontend, /accessibility, react-conventions
backend/      # Backend team — /review-api, /database-migration, api-conventions
infra/        # Infra team — /review-terraform, /incident-runbook, terraform-conventions

Each team folder contains a full canonical layout (skills/, rules/, workflows/). With Forge, each team points their source at the repo with a subdirectory filter or uses include/exclude globs for fine-grained control:

# Frontend team's forge config
sources:
  - name: shared
    type: git
    url: https://github.com/your-org/forge-skills-starter.git
    branch: example/monorepo
    subdirectory: shared

  - name: frontend
    type: git
    url: https://github.com/your-org/forge-skills-starter.git
    branch: example/monorepo
    subdirectory: frontend

Using with Forge

Fork the starter, build your team’s resources, then deliver with Forge:

forge init
# → Select "git" source → paste your fork's URL

forge sync
# → Skills placed for each AI tool automatically

Forge auto-detects the canonical layout and handles format transforms per assistant.