Easy Workflow Harness: A Declarative Workflow Orchestration Framework for Claude Code

Easy Workflow Harness (EWH) is a declarative workflow orchestration framework designed specifically for Claude Code, aiming to address the core problem of evolving AI-assisted programming from simple Q&A mode to a structured, repeatable, and auditable engineering process. Through workflow orchestration, rule injection, agent management, and multi-stage task scheduling mechanisms, the framework enables structured management of complex software development tasks, supports flexible configurations from out-of-the-box use to deep customization, and can be extended to various knowledge work scenarios.

Traditional AI programming assistants mostly adopt an 'ask-and-answer' interaction mode, lacking structured management of the entire development process. EWH introduces the concept of workflow orchestration, decomposing complex tasks into a sequence of predefined steps, each executed by a specific agent and following clear rule constraints. The goal is to make AI-assisted programming a repeatable and auditable engineering process.

EWH defines a workflow as an ordered sequence of steps, each containing four key elements: executing agent, control gate, rule injection, and task description. The framework has six built-in core workflows: init (project initialization), add-feature (feature development closed loop), refine-feature (code optimization), fact-check (document verification), knowledge-update (knowledge base maintenance), clean-up (project health check), covering the entire software development process.

Rules are the core of EWH quality control, divided into four categories: coding (coding standards), testing (testing requirements), review (review criteria), and knowledge (document maintenance), supporting custom extensions. Agents include four predefined roles: coder (code implementation), reviewer (code review), tester (test execution), compliance (compliance audit), and their capabilities can be extended through configuration.

The scheduler is the core of EWH, responsible for coordinating the execution of the entire workflow: reading workflow definitions → loading rule sets → parsing agent configurations → constructing prompts → generating agent conversations → collecting results → compliance checks → gate control (pause or auto-continue). This mechanism ensures tasks are decomposed into manageable and traceable steps, each with a clear responsible entity and quality standards.

EWH supports three layers of configuration: 1. Zero-configuration layer (run commands directly, interactive information completion); 2. Initialization configuration layer (generate CLAUDE.md via init to record project information); 3. Deep customization layer (customize agents, rules, and workflows through the .claude directory), balancing convenience and customization needs.

Custom workflows use YAML metadata + Markdown format, defining names, trigger commands, and step sequences (including agents, gates, rules, etc.). Custom rules also use YAML metadata, specifying applicable scenarios, injection targets, compliance check commands, etc. Failure of key rules will force the process to pause.

EWH embodies the evolution of AI applications from prompt engineering to structured process orchestration. Its core concepts include separation of responsibilities (agent division), built-in quality (embedded rules), auditability (step records), and progressive enhancement (capability expansion). This architectural pattern can be extended to knowledge work scenarios such as document writing, data analysis, and content review.