dify-workflow

Name: dify-workflow
Author: Akabane71

17stars4forks

How to install dify-workflow?

akabane71/dify-workflow-cli/dify-workflow

$npx -y skills add Akabane71/dify-workflow-cli --skill dify-workflow

Installs into the current project.

Summary

Build, edit, validate, and export Dify AI workflow DSL files using the dify-workflow CLI. Use when: creating Dify workflows/chatflows/chat/agent/completion apps; adding nodes and edges to workflows; configuring LLM/tool/code/if-else/question-classifier nodes; validating DSL befor

Files · 1

View on GitHub

SKILL.md

1	# Dify Workflow CLI Skill
2
3	Build, edit, validate, layout, and export Dify DSL files entirely from the command line.
4	The CLI is installed as `dify-workflow` (or `dify-workflow.exe` on Windows).
5
6	If the CLI is not yet installed in the current environment, use [PowerShell install script](./scripts/install.ps1) on Windows or [shell install script](./scripts/install.sh) on macOS/Linux.
7
8	## When to Use
9
10	- User asks to create a Dify workflow, chatflow, chat app, agent, or completion app
11	- User asks to add/remove/update nodes or edges in a Dify YAML
12	- User asks to validate a Dify DSL file before import
13	- User asks to auto-layout node positions
14	- User asks to generate a Mermaid diagram of a workflow
15	- User asks to troubleshoot why a Dify import shows disconnected nodes or errors
16
17	## Key Constraints
18
19	- DAG only: Dify workflows must be directed acyclic graphs. No cycles allowed — Dify's frontend runs `getCycleEdges()` and silently removes all edges between cycle nodes, causing disconnections.
20	- Variable references: Use `{{#node_id.variable#}}` syntax. Tool nodes output `text`/`files`/`json` (NOT `result`). LLM nodes output `text`.
21	- Node positioning: Dify uses screen coordinates (origin top-left, X→right, Y→down). Use `layout` command to auto-arrange.
22	- PowerShell JSON: On Windows PowerShell, avoid inline JSON with `--data`. Use `--data-file` with UTF-8 (no BOM) files instead.
23
24	## Dify App Modes
25
26	Dify supports 5 app modes, split into two architectures:
27
28	### Graph-based modes (use `workflow.nodes` + `workflow.edges`)
29	\| Mode \| DSL `app.mode` \| Description \|
30	\|------\|----------------\|-------------\|
31	\| Workflow \| `workflow` \| Single-run DAG execution, no conversation. Start → nodes → End. \|
32	\| Chatflow \| `advanced-chat` \| Multi-turn conversation with graph canvas. Uses Answer nodes instead of End. \|
33
34	### Config-based modes (use `model_config` section)
35	\| Mode \| DSL `app.mode` \| Description \|
36	\|------\|----------------\|-------------\|
37	\| Chat \| `chat` \| Simple LLM chat with optional knowledge retrieval. No graph. \|
38	\| Agent \| `agent-chat` \| Chat + tool calling. `agent_mode.enabled=true` with strategy (function_call/react). \|
39	\| Completion \| `completion` \| Single-turn text generation. Supports `more_like_this`, requires `dataset_query_variable` for knowledge retrieval. \|
40
41	### Validation coverage by mode
42	- workflow / chatflow: Full graph validation (node data, edges, cycles, connectivity, frontend crash prevention, publish checklist)
43	- chat / agent-chat / completion: `model_config` validation (model, prompt, variables, dataset, agent_mode, features) via `model_config_validators/` package
44
45	## Procedure: Create a Workflow from Scratch
46
47	Before editing a workflow, prefer this execution order:
48
49	1. Ensure the CLI is installed
50	2. Create or open a DSL file
51	3. Add/update nodes and edges
52	4. Run `validate`
53	5. Run `checklist`
54	6. Run `layout`
55	7. Export or inspect with Mermaid
56
57	If the user asks for a complex node payload, prefer using templates in [assets](./assets/).
58
59	### Step 1: Create base file
60
61	```bash
62	dify-workflow create --mode workflow --name "My Workflow" -o workflow.yaml
63	# Templates: minimal (default), llm, if-else
64	# Modes: workflow, chatflow, chat, agent, completion
65	```
66
67	### Step 2: Add nodes
68
69	```bash
70	# Add nodes one at a time
71	dify-workflow edit add-node -f workflow.yaml --type llm --title "GPT Node" --id my_llm
72
73	# With data (prefer --data-file on Windows to avoid escaping issues)
74	dify-workflow edit add-node -f workflow.yaml --type code --title "Process" --id processor
75	dify-workflow edit update-node -f workflow.yaml --id my_llm --data-file ./assets/llm-node-config.json
76	```
77
78	### Step 3: Add edges
79
80	```bash
81	dify-workflow edit add-edge -f workflow.yaml --source start_node --target my_llm
82	dify-workflow edit add-edge -f workflow.yaml --source my_llm --target end_node
83
84	# For branching nodes (if-else, question-classifier), specify --source-handle
85	dify-workflow edit add-edge -f workflow.yaml -s ifelse_node -t happy_path --source-handle true
86	dify-workflow edit add-edge -f workflow.yaml -s ifelse_node -t sad_path --source-handle false
87	```
88
89	### Step 4: Update node data
90
91	```bash
92	dify-workflow edit update-node -f workflow.yaml --id my_llm \
93	-d '{"model": {"provider": "openai", "name": "gpt-4o-mini", "mode": "chat", "completion_params": {"temperature": 0.7}}}'
94
95	# Or from file (recommended on Windows)
96	dify-workflow edit update-node -f workfl

People who liked this also viewed

6 picks

Type

dify-workflow

17stars4forks

View on GitHub

How to install dify-workflow?

akabane71/dify-workflow-cli/dify-workflow

$npx -y skills add Akabane71/dify-workflow-cli --skill dify-workflow

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# Dify Workflow CLI Skill
2
3	Build, edit, validate, layout, and export Dify DSL files entirely from the command line.
4	The CLI is installed as `dify-workflow` (or `dify-workflow.exe` on Windows).
5
6	If the CLI is not yet installed in the current environment, use [PowerShell install script](./scripts/install.ps1) on Windows or [shell install script](./scripts/install.sh) on macOS/Linux.
7
8	## When to Use
9
10	- User asks to create a Dify workflow, chatflow, chat app, agent, or completion app
11	- User asks to add/remove/update nodes or edges in a Dify YAML
12	- User asks to validate a Dify DSL file before import
13	- User asks to auto-layout node positions
14	- User asks to generate a Mermaid diagram of a workflow
15	- User asks to troubleshoot why a Dify import shows disconnected nodes or errors
16
17	## Key Constraints
18
19	- DAG only: Dify workflows must be directed acyclic graphs. No cycles allowed — Dify's frontend runs `getCycleEdges()` and silently removes all edges between cycle nodes, causing disconnections.
20	- Variable references: Use `{{#node_id.variable#}}` syntax. Tool nodes output `text`/`files`/`json` (NOT `result`). LLM nodes output `text`.
21	- Node positioning: Dify uses screen coordinates (origin top-left, X→right, Y→down). Use `layout` command to auto-arrange.
22	- PowerShell JSON: On Windows PowerShell, avoid inline JSON with `--data`. Use `--data-file` with UTF-8 (no BOM) files instead.
23
24	## Dify App Modes
25
26	Dify supports 5 app modes, split into two architectures:
27
28	### Graph-based modes (use `workflow.nodes` + `workflow.edges`)
29	\| Mode \| DSL `app.mode` \| Description \|
30	\|------\|----------------\|-------------\|
31	\| Workflow \| `workflow` \| Single-run DAG execution, no conversation. Start → nodes → End. \|
32	\| Chatflow \| `advanced-chat` \| Multi-turn conversation with graph canvas. Uses Answer nodes instead of End. \|
33
34	### Config-based modes (use `model_config` section)
35	\| Mode \| DSL `app.mode` \| Description \|
36	\|------\|----------------\|-------------\|
37	\| Chat \| `chat` \| Simple LLM chat with optional knowledge retrieval. No graph. \|
38	\| Agent \| `agent-chat` \| Chat + tool calling. `agent_mode.enabled=true` with strategy (function_call/react). \|
39	\| Completion \| `completion` \| Single-turn text generation. Supports `more_like_this`, requires `dataset_query_variable` for knowledge retrieval. \|
40
41	### Validation coverage by mode
42	- workflow / chatflow: Full graph validation (node data, edges, cycles, connectivity, frontend crash prevention, publish checklist)
43	- chat / agent-chat / completion: `model_config` validation (model, prompt, variables, dataset, agent_mode, features) via `model_config_validators/` package
44
45	## Procedure: Create a Workflow from Scratch
46
47	Before editing a workflow, prefer this execution order:
48
49	1. Ensure the CLI is installed
50	2. Create or open a DSL file
51	3. Add/update nodes and edges
52	4. Run `validate`
53	5. Run `checklist`
54	6. Run `layout`
55	7. Export or inspect with Mermaid
56
57	If the user asks for a complex node payload, prefer using templates in [assets](./assets/).
58
59	### Step 1: Create base file
60
61	```bash
62	dify-workflow create --mode workflow --name "My Workflow" -o workflow.yaml
63	# Templates: minimal (default), llm, if-else
64	# Modes: workflow, chatflow, chat, agent, completion
65	```
66
67	### Step 2: Add nodes
68
69	```bash
70	# Add nodes one at a time
71	dify-workflow edit add-node -f workflow.yaml --type llm --title "GPT Node" --id my_llm
72
73	# With data (prefer --data-file on Windows to avoid escaping issues)
74	dify-workflow edit add-node -f workflow.yaml --type code --title "Process" --id processor
75	dify-workflow edit update-node -f workflow.yaml --id my_llm --data-file ./assets/llm-node-config.json
76	```
77
78	### Step 3: Add edges
79
80	```bash
81	dify-workflow edit add-edge -f workflow.yaml --source start_node --target my_llm
82	dify-workflow edit add-edge -f workflow.yaml --source my_llm --target end_node
83
84	# For branching nodes (if-else, question-classifier), specify --source-handle
85	dify-workflow edit add-edge -f workflow.yaml -s ifelse_node -t happy_path --source-handle true
86	dify-workflow edit add-edge -f workflow.yaml -s ifelse_node -t sad_path --source-handle false
87	```
88
89	### Step 4: Update node data
90
91	```bash
92	dify-workflow edit update-node -f workflow.yaml --id my_llm \
93	-d '{"model": {"provider": "openai", "name": "gpt-4o-mini", "mode": "chat", "completion_params": {"temperature": 0.7}}}'
94
95	# Or from file (recommended on Windows)
96	dify-workflow edit update-node -f workfl

People who liked this also viewed

6 picks

Type