nw-agent-creation-workflow

A phase-driven agent creation workflow orchestrates AI agent development through five sequential stages: ANALYZE, DESIGN, CREATE, VALIDATE, and REFINE. Each stage has explicit inputs, outputs, and quality gates to ensure safe, repeatable deployment and iterative improvement.

You validate AI agents before deployment by applying a 14-point checklist during the VALIDATE phase of the workflow. This governance step checks frontmatter compliance, artifact organization, and skill-loading directives to ensure the agent meets all quality and safety requirements.

The best way to structure AI agent development and avoid duplication is using a deterministic, phase-sequenced workflow. It enforces architecture design and artifact organization rules that reduce overlap with existing agents before you create new ones.

No specific dependencies are required to use this structured agent design process. The workflow is self-contained and relies on its own internal phase sequencing, frontmatter compliance rules, and validation gates rather than external libraries or components.

You should not use a deterministic workflow for AI agent creation when your project requires rapid, informal prototyping without strict governance. The explicit phase gates, frontmatter compliance, and validation checkpoints are designed for rigorous engineering, not ad-hoc builds.