/dify-workflow/agent-harness/DIFY_WORKFLOW.md

Dify Workflow Harness Architecture

Overview

This harness adds cli-anything-dify-workflow, a CLI-Anything wrapper around the existing dify-workflow CLI from the open-source dify-ai-workflow-tools project.

The upstream CLI already provides the real workflow authoring engine for Dify DSL files. This harness focuses on:

• CLI-Anything packaging under the shared cli_anything namespace
• AI-discoverable SKILL.md
• unified REPL skin
• registry integration for CLI-Hub
• tests that verify wrapper discovery and forwarding behavior

Interaction Model

AI Agent
  -> cli-anything-dify-workflow
     -> installed dify-workflow CLI / dify_workflow Python package
        -> local Dify YAML/JSON DSL files

Why a Wrapper Harness

The upstream project is already a mature CLI and does not need to be rewritten inside CLI-Anything. Following existing wrapper patterns in this repository, this harness exposes the upstream CLI through a CLI-Anything package so agents can discover it from CLI-Hub and load its skill metadata.

Requirements

• Python 3.10+
• Upstream Dify workflow CLI installed separately:
  • recommended: install from GitHub
  • optional: install from PyPI if published later

Command Surface

The wrapper exposes the upstream command groups:

• guide
• list-node-types
• create
• inspect
• validate
• checklist
• edit <subcommand>
• config <subcommand>
• export
• import
• diff
• layout

Testing Strategy

• test_core.py: backend discovery, command building, wrapper metadata
• test_full_e2e.py: wrapper subprocess smoke tests and workflow lifecycle forwarding when the upstream CLI is installed