Skip to content

agentsope/SkillAlchemy / skills/agentsop-dify

agentsop-dify

Verified

SOP for building LLM applications on Dify — visual workflow + chatflow + agent + RAG knowledge base + plugin marketplace + observability, self-hostable. Use when shipping LLM apps fast with a "no-code to pro-code" gradient, especially when non-engineers need to co-author the flow.

intermediate/operation_candidates.json

View on GitHub
{
  "skill": "dify-sop",
  "version": "1.0.0",
  "framework_version": "1.14.x (May 2026)",
  "research_mode": "DEEP",
  "dimensions_covered": ["R1", "R2", "R3", "R4", "R5"],
  "core_primitives": [
    {
      "name": "App",
      "variants": ["Chatbot (legacy)", "Agent (legacy)", "Text Generator (legacy)", "Workflow", "Chatflow"],
      "recommendation": "Workflow / Chatflow for long-lived projects; legacy types only for simplicity",
      "source": "docs.dify.ai/en/use-dify/getting-started/key-concepts"
    },
    {
      "name": "Graph Engine (graphon)",
      "description": "Unified DAG runtime under all app types; v1.9+ queue-based scheduling",
      "source": "github.com/langgenius/dify/discussions/26138"
    },
    {
      "name": "Knowledge Base",
      "description": "RAG pipeline: ingest → chunk → embed → index → retrieve → rerank",
      "creation_paths": ["Quick create", "Knowledge Pipeline (v1.9+)", "External integration"],
      "vector_stores": ["Qdrant (recommended)", "Weaviate", "Milvus", "Pinecone", "PGVector", "TiDB", "Chroma", "+6 more"],
      "source": "deepwiki.com/langgenius/dify-docs/8.3-vector-database-configuration"
    },
    {
      "name": "Plugin System",
      "types": ["Models", "Tools", "Agent Strategies", "Extensions", "Bundles"],
      "introduced": "v1.0",
      "source": "dify.ai/blog/introducing-dify-plugins"
    },
    {
      "name": "Monitoring",
      "integrations": ["LangSmith", "Langfuse", "Arize Phoenix", "Opik"],
      "trace_types": ["Workflows", "Messages", "Moderation", "Suggested Questions", "Dataset Retrieval", "Tools", "Generated Names"],
      "source": "docs.dify.ai/guides/monitoring/integrate-external-ops-tools"
    }
  ],
  "node_taxonomy": {
    "flow": ["Start", "End", "Answer"],
    "llm": ["LLM", "Question Classifier", "Parameter Extractor"],
    "rag": ["Knowledge Retrieval"],
    "logic": ["IF/ELSE", "Iteration", "Loop", "Variable Assigner", "Variable Aggregator"],
    "code": ["Code (Python/Node.js)", "Template (Jinja2)"],
    "external": ["HTTP Request", "Tool", "Agent Node"],
    "data": ["List Operator", "Document Extractor"]
  },
  "decision_points": [
    {
      "id": "DP1",
      "question": "Conversation context needed?",
      "options": ["Chatflow (complex)", "Chatbot/Agent (simple)", "Workflow (no context, complex)", "Text-Gen (no context, simple)"],
      "rule": "if multi-turn memory required → Chatflow family else Workflow family"
    },
    {
      "id": "DP2",
      "question": "Deterministic logic or LLM autonomous decisions?",
      "options": ["Explicit Workflow", "Workflow + Agent Node (hybrid, recommended)", "Pure Agent app"],
      "rule": "hybrid wins for most production cases; pure Agent only for simple dialogue"
    },
    {
      "id": "DP3",
      "question": "Cloud or self-host?",
      "factors": ["compliance", "volume", "team size", "ops capacity", "data residency"],
      "default": "Cloud Pro ($59) for teams ≤3; self-host for compliance / heavy volume / air-gapped"
    },
    {
      "id": "DP4",
      "question": "Dify built-in KB or external vector store?",
      "rule": "single app / single dataset → built-in; multi-app share / extreme RAG / >100M chunks → external"
    },
    {
      "id": "DP5",
      "question": "Tool extension method?",
      "options": ["HTTP Request node (one-off)", "Tool plugin (reusable)", "Code node (custom logic)", "Agent Strategy plugin (reasoning loop)"]
    },
    {
      "id": "DP6",
      "question": "Workflow getting unwieldy — what to do?",
      "thresholds": {
        "<20 nodes flat": "stay in Dify, rename + Note",
        "20-40 nodes with sub-procedures": "split into sub-workflows as tools",
        ">40 nodes or >100-line Code": "externalize logic to microservice"
      }
    }
  ],
  "dilemma_cases": [
    {
      "id": "C1",
      "title": "Workflow unmaintainable — split or rewrite?",
      "sources": ["github.com/langgenius/dify/issues/28245", "memo.d.foundation/breakdown/dify"]
    },
    {
      "id": "C2",
      "title": "Visual ceiling hit — Dify or LangGraph/code?",
      "sources": ["github.com/langgenius/dify/issues/21455", "github.com/langgenius/dify/issues/24791", "blog.elest.io/dify-vs-langflow-vs-flowise"]
    },
    {
      "id": "C3",
      "title": "Self-host vs Cloud — when is ops cost worth it?",
      "sources": ["architjn.com/blog/dify-cloud-pricing", "github.com/langgenius/dify/discussions/32254"]
    },
    {
      "id": "C4",
      "title": "Dify Knowledge vs separate vector store?",
      "sources": ["deepwiki.com/langgenius/dify-docs/8.3-vector-database-configuration", "sider.ai/blog/ai-tools/dify-vs-ragflow", "memo.d.foundation/breakdown/dify"]
    },
    {
      "id": "C5",
      "title": "Agent app vs Workflow+Agent Node vs explicit Workflow?",
      "sources": ["zediot.com/blog/dify-difference-between-agent-and-workflow", "dify.ai/blog/dify-agent-node-introduction-when-workflows-learn-autonomous-reasoning"]
    },
    {
      "id": "C6",
      "title": "Visual builder feels constrained — is Dify still right?",
      "sources": ["zhuanlan.zhihu.com/p/1947389040702781389"]
    }
  ],
  "anti_patterns": [
    "Mega-workflow with 30+ nodes (canvas lag, hard to debug)",
    "Code node carrying 100+ lines of business logic",
    "Long-term project built on legacy Chatbot/Agent app types",
    "No DSL exported to git for version control",
    "No external LLMOps wired (Langfuse/Phoenix/LangSmith)",
    "Workspace-shared model API keys causing cost-accounting issues",
    "Skipping retrieval testing on knowledge base",
    "Long agent streams (>120s) without timeout tuning",
    "Beta upgrade without backup (v1.9 destroys beta KB pipelines)",
    "Direct container port exposed to internet (no reverse proxy)",
    "Container-internal PostgreSQL in production",
    "Agent app for tasks with predictable path"
  ],
  "performance_boundaries": {
    "qps_per_pod": "~10 (1 CPU 2GB)",
    "plugin_daemon_timeout_default": "300s",
    "agent_stream_with_logs_disconnect": "120s",
    "workflow_max_execution_time": "configurable env var",
    "cloud_storage_quota": "5MB Sandbox / 5GB Pro / 20GB Team"
  },
  "not_for": [
    "Model training / fine-tuning",
    "General workflow automation with mostly non-AI integrations",
    "Extreme RAG quality (KG-RAG, deep parsing) — use RAGFlow",
    "Pause-and-wait-for-user / approval flows — use LangGraph",
    "High-frequency real-time inference router — use vLLM + custom gateway",
    "Heavy-state agent runtime with time-travel — use LangGraph",
    "Sub-second latency endpoints",
    "Code node as microservice"
  ],
  "ecosystem_positioning": {
    "vs_Flowise": "Dify = production SaaS; Flowise = minimal chatbot editor (1GB RAM)",
    "vs_LangFlow": "Dify = own framework, multi-tenant; LangFlow = LangChain-coupled, exports code",
    "vs_Coze": "Coze = ByteDance hosted only + Feishu/Lark; Dify = open source, self-host",
    "vs_RAGFlow": "Dify = platform with RAG; RAGFlow = specialized RAG engine with thin UI",
    "vs_n8n": "Dify = AI-native; n8n = general automation with 400+ integrations and AI as plugin",
    "vs_LangGraph": "Dify = visual platform; LangGraph = code framework with rich state machine",
    "vs_LlamaIndex": "Dify = app platform; LlamaIndex = data/RAG framework. Often combined via External KB",
    "vs_CrewAI": "Dify = workflow + visual; CrewAI = role-based multi-agent in code. Often combined"
  },
  "surprising_findings": [
    "Dify's 'low-code' claim is contested — Chinese community openly argues that to build a real agent you still need Python (zhuanlan.zhihu.com/p/1947389040702781389), making Dify's true value 'scaffolding compression' not 'no-code'",
    "Pause-and-wait-for-user is a hard not-supported (issue #21455 closed 'not planned') — surprising for a 'production' platform claiming agent workflows. Forces hybrid LangGraph architecture for approval flows",
    "Per-node DB queries cap throughput at ~10 QPS per pod regardless of model latency — the bottleneck is Dify's own engine, not the LLM (memo.d.foundation/breakdown/dify)",
    "All Dify editions (Community/Premium/Enterprise) use the SAME Docker images — Enterprise features are unlocked purely by env-var flags pointing at LangGenius's API (discussions/32254)",
    "The platform is honest about its limits in its own GitHub discussions — team transparently lists bottlenecks and roadmap, which is unusual for a vendor-backed open source project"
  ],
  "key_sources": [
    "https://github.com/langgenius/dify",
    "https://docs.dify.ai/",
    "https://marketplace.dify.ai/",
    "https://dify.ai/blog/",
    "https://memo.d.foundation/breakdown/dify",
    "https://blog.elest.io/dify-vs-langflow-vs-flowise-which-open-source-llm-app-builder-actually-ships-to-production/",
    "https://jimmysong.io/blog/open-source-ai-agent-workflow-comparison/",
    "https://sider.ai/blog/ai-tools/dify-vs-ragflow-which-rag-platform-should-you-build-on-in-2025",
    "https://hellodify.com/en/docs/workflow/workflow-chatflow-difference",
    "https://zediot.com/blog/dify-difference-between-agent-and-workflow/",
    "https://learnwithparam.com/blog/batteries-included-rag-platforms-dify-ragflow-onyx",
    "https://zhuanlan.zhihu.com/p/1947389040702781389",
    "https://github.com/langgenius/dify/issues/21455",
    "https://github.com/langgenius/dify/issues/28245",
    "https://github.com/langgenius/dify/discussions/26138",
    "https://github.com/langgenius/dify/discussions/27053",
    "https://github.com/langgenius/dify/discussions/32254",
    "https://architjn.com/blog/dify-cloud-pricing-plans-free-tier-when-to-self-host"
  ]
}
reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness.Stars: 322kUpdated Claude CodeCodexVerifiedInstall
dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependenciesStars: 322kUpdated Claude CodeCodexVerifiedInstall
executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is availableStars: 322kUpdated Claude CodeCodexVerifiedInstall