Dify Master SKILL
A modular, progressively disclosed Dify development skill system
Quick Navigation
What is Dify
Dify is an open-source LLMOps platform that provides visual workflow orchestration and a powerful plugin system, helping developers quickly build and deploy AI applications.
Core Features
- Visual Workflows: Drag-and-drop orchestration, build complex AI apps without coding
- Plugin System: Supports multiple plugin types: tools, models, data sources
- Production Ready: Complete support from local development to large-scale distributed deployment
- Open Source Ecosystem: 114K+ GitHub Stars, active community, and rich plugin marketplace
Technical Architecture
- Beehive Architecture: Modular, extensible system design
- Queue-Driven Graph Engine: Event-driven, parallel execution, high performance
- Multi-Runtime Support: Local, Debug, and Serverless operating modes
- MCP Protocol Integration: Standardized integration with AI assistants like Claude
SKILL System Overview
Dify Master SKILL adopts a modular design with 6 core sub-skills covering the complete development-to-deployment pipeline.
dify-master (main entry)
├── 01-workflow # Workflow design
├── 02-plugin # Plugin development
├── 03-performance # Performance optimization
├── 04-security # Security and deployment
├── 05-integration # Integration and extension
└── 06-reference # Reference materials
Design Philosophy
- Modular: Each sub-skill is independent and complete, usable on its own
- Progressive Disclosure: From overview to details, go deeper as needed
- Context Optimized: Minimize the amount of content loaded per session
- Practical Oriented: Provide ready-to-use code and configurations
Quick Start
Scenario 1: Create Your First Workflow
# Simple Q&A workflow
workflow:
graph:
nodes:
- data:
type: start
variables: []
id: start
- data:
type: llm
model:
provider: openai
name: gpt-4
prompt_template:
- role: system
text: "You are an assistant"
- role: user
text: "{{#sys.query#}}"
id: llm
- data:
type: answer
answer: "{{#llm.text#}}"
id: answer
edges:
- source: start
target: llm
- source: llm
target: answer
Next step: Learn 01-workflow SKILL for more node types and design patterns
Scenario 2: Develop Your First Plugin
# Simple tool plugin
from typing import Any, Generator
from dify_plugin.entities.tool import ToolInvokeMessage
from dify_plugin import Tool
class MyTool(Tool):
def _invoke(
self,
tool_parameters: dict[str, Any]
) -> Generator[ToolInvokeMessage, None, None]:
# Get parameters
input_text = tool_parameters.get("text", "")
# Processing logic
result = input_text.upper()
# Return result
yield self.create_text_message(result)
Next step: Learn 02-plugin SKILL for the complete plugin development workflow
Scenario 3: Optimize Workflow Performance
# Use parallel branches to boost performance
- data:
type: parallel
branches:
- nodes: [tool1, process1]
- nodes: [tool2, process2]
- nodes: [tool3, process3]
id: parallel_execution
Next step: Learn 03-performance SKILL for more optimization techniques
Sub-SKILL Index
01. Workflow Design SKILL
Use Cases: Design and develop Dify workflows
Trigger Keywords: node, LLM, Agent, workflow, variable, knowledge retrieval, HTTP Request, Code, If-Else, Iteration, template, DSL, YAML
Core Content:
- 10+ node types explained (Start, LLM, Agent, Knowledge Retrieval, Code, HTTP Request, Tool, If-Else, Iteration, Answer)
- 5 design patterns (Sequential Pipeline, Conditional Branching, Iterative Processing, Knowledge Retrieval, Multi-Source Aggregation)
- Ready-to-use workflow templates
- Variable system and reference syntax
- Best practices and common issues
Quick Access: 01-workflow/SKILL.md
Recommended for: Workflow designers, AI application developers, product managers
02. Plugin Development SKILL
Use Cases: Develop custom tool, model, and data source plugins
Trigger Keywords: plugin, Tool, Provider, development, manifest, OAuth, data source, model, extension
Core Content:
- 4 plugin types (Tool, Model, Datasource, Extension)
- Complete development workflow (initialize, implement, test, package, publish)
- Standard project structure and configuration files
- Provider and Tool development guides
- Plugin templates and code examples
- Testing strategies and debugging tips
Quick Access: 02-plugin/SKILL.md
Recommended for: Plugin developers, backend engineers, tool integrators
03. Performance Optimization SKILL
Use Cases: Optimize execution performance of workflows and plugins
Trigger Keywords: performance, optimization, cache, parallel, latency, throughput, monitoring, benchmark
Core Content:
- Workflow optimization (graph structure, parallel execution, variable pool)
- LLM optimization (model selection, prompt engineering, caching, structured output)
- Plugin optimization (runtime selection, resource limits, async processing)
- Monitoring and analysis tools
- Performance benchmarks
Quick Access: 03-performance/SKILL.md
Recommended for: Performance engineers, architects, operations staff
04. Security and Deployment SKILL
Use Cases: Production environment deployment and security hardening
Trigger Keywords: security, deployment, Docker, Kubernetes, SSL, authentication, authorization, sandbox, SSRF
Core Content:
- Security practices (SSRF protection, sandbox isolation, data encryption, authentication/authorization)
- Deployment guides (Docker Compose, Kubernetes, Serverless, high availability)
- Operations practices (monitoring, logging, backup, troubleshooting)
- Environment configuration and resource planning
- Best practices and case studies
Quick Access: 04-security/SKILL.md
Recommended for: DevOps engineers, security engineers, system administrators
05. Integration and Extension SKILL
Use Cases: Integrate with external systems and services
Trigger Keywords: integration, API, Webhook, MCP, Server, Client, database, Redis, message queue, RabbitMQ, Kafka, S3
Core Content:
- API integration (REST API, Webhook, Streaming)
- MCP protocol (MCP Server, MCP Client, bidirectional communication)
- External services (databases, storage, message queues)
- Integration case studies and best practices
- Troubleshooting and debugging
Quick Access: 05-integration/SKILL.md
Recommended for: Integration engineers, full-stack developers, architects
06. Reference SKILL
Use Cases: Look up syntax, API, CLI, and other reference information
Trigger Keywords: reference, API, SDK, CLI, DSL, YAML, command, documentation, syntax
Core Content:
- DSL syntax reference (complete YAML structure and field descriptions)
- API reference (RESTful API interface documentation)
- SDK reference (Python SDK classes and methods)
- CLI reference (dify command line tool)
- Common questions and solutions
- Learning resources and community links
Quick Access: 06-reference/SKILL.md
Recommended for: All Dify developers
Common References
Variable Reference Syntax
# System variables
{{#sys.query#}} # User query
{{#sys.files#}} # Uploaded files
{{#sys.conversation_id#}} # Conversation ID
# Node output
{{#nodeID.outputField#}} # Reference node output
{{#llm.text#}} # LLM text output
{{#knowledge.result#}} # Knowledge retrieval result
# Conversation variables
{{#conversation.variableName#}} # Conversation-level variable
# Iteration variables
{{#item#}} # Current iteration item
{{#index#}} # Current index
Common Node Configurations
# LLM node
- data:
type: llm
model:
provider: openai
name: gpt-4
completion_params:
temperature: 0.7
max_tokens: 2000
prompt_template:
- role: system
text: "System prompt"
- role: user
text: "{{#sys.query#}}"
context:
enabled: true
variable_selector: ["knowledge", "result"]
# Agent node
- data:
type: agent
agent_parameters:
instruction:
value: "Call tools based on user requirements"
model:
value:
provider: openai
model: gpt-4
query:
value: "{{#sys.query#}}"
tools:
value:
- enabled: true
provider_name: time
tool_name: current_time
type: builtin
agent_strategy_name: function_calling
# Knowledge retrieval node
- data:
type: knowledge-retrieval
dataset_ids: ["dataset-id"]
query_variable_selector: ["start", "sys.query"]
retrieval_mode: single
single_retrieval_config:
model:
provider: openai
name: text-embedding-3-small
top_k: 3
score_threshold: 0.7
Plugin Configuration Template
# manifest.yaml
version: 0.0.1
type: plugin
author: your-name
name: plugin-name
label:
en_US: Plugin Name
zh_Hans: Plugin Name
description:
en_US: Plugin description
zh_Hans: Plugin description
icon: icon.svg
resource:
memory: 268435456 # 256MB
plugins:
tools:
- provider/provider.yaml
meta:
version: 0.0.1
arch: [amd64, arm64]
runner:
language: python
version: "3.12"
entrypoint: main
tags:
- utilities
Common CLI Commands
# Plugin development
dify plugin init # Initialize a plugin project
dify plugin run # Debug plugin locally
dify plugin package # Package plugin
dify plugin publish # Publish plugin
# Workflow management
dify workflow export <id> # Export a workflow
dify workflow import <file> # Import a workflow
dify workflow validate <file> # Validate a workflow
# Environment management
dify env list # List environment variables
dify env set <key> <value> # Set environment variable
dify env get <key> # Get environment variable
Learning Path
Beginner Path (0-1 Week)
- Read this SKILL to understand the Dify landscape
- Learn 01-workflow SKILL to master workflow fundamentals
- Create your first workflow using a template
- Test and debug on the Dify platform
Intermediate Path (1-4 Weeks)
- Deep-dive into 01-workflow SKILL design patterns
- Learn 02-plugin SKILL to develop custom plugins
- Learn 03-performance SKILL to optimize performance
- Build and deploy a real project
Expert Path (1-3 Months)
- Master all sub-skill content
- Learn 04-security SKILL for production deployment
- Learn 05-integration SKILL to integrate external systems
- Contribute to the community and share best practices
Getting Help
Official Resources
- Documentation: https://docs.dify.ai/
- GitHub: https://github.com/langgenius/dify
- Plugin Marketplace: https://marketplace.dify.ai/
- Official Blog: https://dify.ai/blog/
Community Resources
- Discord: https://discord.gg/FngNHpbcY7
- Reddit: https://reddit.com/r/difyai
- Chinese Community: https://github.com/langgenius/dify/discussions
SKILL Support
- Architecture Document: ARCHITECTURE.md
- Issue Feedback: Submit via GitHub Issues
- Contribution Guide: Help improve and extend SKILL content
Version Information
- Current Version: v1.1.0
- Last Updated: 2026-03-05
- Compatible Dify Versions: 0.15.x - 1.9.x
- Update Content: Completed 05-integration and 06-reference SKILLs, added on-demand loading mechanism
- Next Version Plan: v1.2.0 (add more templates and case studies)
License
This SKILL is open-sourced under the MIT License. You are free to use, modify, and distribute it.
Start your Dify development journey! Choose a sub-SKILL to dive deeper, or jump directly into templates to get started quickly.