📝 This document is automatically generated and maintained by the GitHub Manager skill (AI-assisted). Do not edit manually — changes will be overwritten on the next update.
A Claude Code skill that converts natural language descriptions into importable Dify workflow YAML files. Runs a 4-agent pipeline to analyze intent, design architecture, generate DSL, and validate the output.
Dify Assistant is a developer tool for building Dify workflows without manually writing DSL. You describe what you want in plain language, and the system produces a ready-to-import YAML file along with a delivery report and first-run checklist.
The core skill (/dify-workflow) orchestrates four specialized AI agents in sequence:
Intent Analysis → Architecture Design → DSL Generation → Validation & Delivery
- Natural language to workflow — describe inputs, processing steps, and outputs in plain text
- Model selection with recommendations — suggests the best model based on your workflow type (creative writing, RAG, reasoning, etc.)
- Multi-pattern support — linear chains, parallel branches, RAG retrieval, conditional routing, iteration loops
- Auto-validation — machine schema check + LLM structural check, with auto-fix for common issues
- Delivery report — every generated workflow includes a design rationale document, usage guide, known limitations, and iteration suggestions
- First-run checklist — concrete steps to verify your Dify instance is ready (API keys, model availability, placeholder replacements)
This project runs as a Claude Code skill. Prerequisites:
- Claude Code CLI installed
- Node.js 18+ (for the validation script)
# Clone the repo
git clone https://github.com/hunterhigh/dify-assistant.git
cd dify-assistant
# Install validation script dependencies
cd plugins/dify-workflow/scripts
npm installThen register the skill in Claude Code:
# Copy skill to your Claude Code skills directory
cp plugins/dify-workflow/skill.md ~/.claude/skills/dify-workflow/SKILL.mdIn Claude Code, invoke the skill:
/dify-workflow
The skill will ask you to describe your workflow using the input/processing/output structure, then recommend a model and run the 4-stage pipeline.
Example:
/dify-workflow
Inputs: user provides a product name and key selling points
Processing: generate 3 versions of copywriting in different styles (casual / professional / storytelling)
Output: 3 titles and body texts
Generated files are saved to:
output/{workflow-name}.yaml— importable Dify DSLoutput/{workflow-name}-report.md— delivery report
dify-assistant/
├── dify-knowledge/ # Reference docs for AI agents
│ ├── nodes.md # Dify node type specs and required fields
│ ├── patterns.md # Workflow design patterns (5 patterns)
│ ├── dsl-schema.md # Dify DSL top-level structure
│ ├── models.md # Model → provider mapping and known constraints
│ └── examples/ # Reference YAML examples
│ ├── rag-basic.yaml
│ ├── parallel-branch.yaml
│ └── agent-loop.yaml
├── plugins/
│ └── dify-workflow/
│ ├── skill.md # Main skill orchestration (entry point)
│ ├── prompts/ # Agent instruction files
│ │ ├── intent-analyst.md
│ │ ├── architect.md
│ │ ├── dsl-generator.md
│ │ └── validator.md
│ ├── scripts/
│ │ └── validate-schema.js # Node.js schema validator
│ └── session/ # Temporary files per run (gitignored)
└── output/ # Generated workflow files (gitignored)
| Dependency | Version | Purpose |
|---|---|---|
| Claude Code | latest | Skill runtime |
| Node.js | 18+ | Schema validation script |
| js-yaml | 4.x | YAML parsing in validator |
| Pattern | Description |
|---|---|
| Linear Chain | Sequential nodes: start → process → end |
| Parallel Branch + Aggregation | Multiple independent LLMs running in parallel |
| Basic RAG Q&A | Knowledge retrieval + LLM answer generation |
| Conditional Routing | if-else or question-classifier branching |
| Iteration Loop | Process a list item by item |
MIT