Consilium Fabri: A Workflow Template for Multi-Agent AI Development with Gatekeeping and Artifact-First Principles

Consilium Fabri (Latin for "Craftsman's Advice") is a workflow template for multi-agent AI development, supporting collaborative work between Claude Code, Gemini CLI, and Codex CLI. Its core design principles are "Gatekeeping" (ensuring code quality) and "Artifact-First" (centered on deliverable artifacts), aiming to solve issues like code quality, conflicts, and consistency in multi-agent collaboration, and provide a reliable practice framework.

With the development of AI coding assistants, developers are exploring multi-agent collaboration to tackle complex tasks, but face challenges like code quality, conflicts, and consistency. Single agents have limitations such as context constraints, capability boundaries, error accumulation, and collaboration blind spots; simple multi-agent collaboration can easily lead to chaos (conflicting modifications, inconsistent styles, tracking difficulties, and quality control issues). Therefore, a structured workflow template is needed to standardize AI team collaboration.

Gatekeeping: Set up multi-layer gates (entry, integration, release) to ensure code quality; divide agent permissions (architect, developer, reviewer, tester) to avoid chaos. Artifact-First: Drive development centered on deliverable artifacts (functional specifications, API definitions, test cases, etc.); artifacts serve as contracts between agents (e.g., the architect's design document as input for developers), ensuring clear and traceable collaboration.

Supports collaboration between Claude Code, Gemini CLI, and Codex CLI. Recommended role division: Claude Code (architectural design, complex logic, coordination, review), Gemini CLI (rapid prototyping, repetitive tasks, document generation, test writing), Codex CLI (Git operations, CI/CD configuration, GitHub integration). Typical workflow: Requirement analysis → Rapid prototyping → Code review → Test supplement → Integration verification → Document improvement.

Recommended project structure includes directories like .consilium (workflow configuration), artifacts (deliverables), src, tests, etc. Gatekeeping rules are configured declaratively (e.g., syntax checks, unit tests, security scans). Define agent interaction protocols (task assignment, status reporting, artifact handover, conflict resolution).

Applicable to large-scale project development (clear boundaries, consistent quality, traceable changes), legacy system maintenance (divided roles to understand old code, write migration tests, refactor), rapid prototype iteration (multi-solution generation and evaluation, deployment integration), and automated code review (style checks, bug detection, architecture verification).

Adopt incrementally (try on small projects → introduce multi-agents → expand to full workflow); key decisions require human supervision (architectural direction, security operations, business logic); continuous optimization (adjust gatekeeping rules, role division, and best practices based on feedback).