Skip to content

init-command.md

Latest commit

 

History

History
164 lines (108 loc) · 4.08 KB

init-command.md

File metadata and controls

164 lines (108 loc) · 4.08 KB
Raw

workflow init — Template Scaffolding

Create new workflow files from built-in templates.

Usage

# List available templates
workflow init --list

# Create from a template (interactive)
workflow init

# Create from a specific template
workflow init --template rag-qa

# Specify output filename
workflow init --template simple-form -o my-workflow.workflow.yaml

# Overwrite existing file
workflow init --template blank -o existing.workflow.yaml --force

Options

Option Short Description
--template NAME -t Template name to scaffold from
--output PATH -o Output file path (default: {template-name}.workflow.yaml)
--force Overwrite existing file if it exists
--list List all available templates and exit

Available Templates

simple-form

Pattern: structured_input → agent

A structured form collects user input, then an agent processes it. Good starting point for data collection workflows.

Nodes: structured_input, agent

text-to-agent

Pattern: plain_txt_input → agent

The simplest agent workflow: free-text input processed directly by an agent. Minimal starting point.

Nodes: plain_txt_input, agent

document-analysis

Pattern: file_upload → retrieve → llm_call

Upload a document, retrieve related context from a knowledge base, and analyze with an LLM. Classic RAG pipeline for document processing.

Nodes: file_upload, retrieve, llm_call

form-with-review

Pattern: structured_input → llm_call → human_review

Structured input processed by an LLM, then routed to a human reviewer for approval. Good for workflows requiring manual oversight.

Nodes: structured_input, llm_call, human_review

batch-processing

Pattern: plain_txt_input → llm_call → llm_call

Text input processed through a multi-stage LLM pipeline. First LLM processes items, second summarizes results.

Nodes: plain_txt_input, llm_call, llm_call

rag-qa

Pattern: plain_txt_input → retrieve → llm_call

Classic RAG question-answering: user asks a question, relevant docs are retrieved, and an LLM generates a grounded answer.

Nodes: plain_txt_input, retrieve, llm_call

blank

Pattern: plain_txt_input (minimal)

Empty workflow with just an entry/exit node. A minimal starting point to build from scratch.

Nodes: plain_txt_input

Examples

Quick Start

# See what's available
workflow init --list

# Create a RAG Q&A workflow
workflow init --template rag-qa

# Output: rag-qa.workflow.yaml created

Custom Filename

workflow init --template document-analysis -o invoice-processor.workflow.yaml

Overwrite Protection

# First time - succeeds
workflow init --template blank -o test.workflow.yaml

# Second time - fails with error
workflow init --template blank -o test.workflow.yaml
# Error: File already exists: test.workflow.yaml
# Use --force to overwrite.

# Third time with --force - succeeds
workflow init --template blank -o test.workflow.yaml --force

Template Validation

All built-in templates are guaranteed to pass workflow validate:

# Create and validate in one go
workflow init --template rag-qa
workflow validate rag-qa.workflow.yaml
# All checks passed ✓

Template Customization

After scaffolding, customize the generated .workflow.yaml file:

  1. Update placeholder values: Replace YOUR_KNOWLEDGE_BASE_ID, agent IDs, etc.
  2. Adjust parameters: Tune temperature, maxTokens, topK, etc.
  3. Modify structure: Add/remove nodes, change edge connections
  4. Validate changes: Run workflow validate after edits

Integration with Other Commands

# Scaffold → Validate → Push
workflow init --template rag-qa -o my-qa.workflow.yaml
workflow validate my-qa.workflow.yaml
workflow push my-qa.workflow.yaml

Reference

  • Jira Ticket: RAG-948
  • Related Commands: workflow validate
  • Template Location: src/cli/templates/
  • WDF Schema: See docs/validate-command.md for node types and validation rules