osmedeus/test/testdata/full-field-workflows/module-llm-comprehensive.yaml

# Comprehensive example demonstrating all LLM step fields
kind: module
name: module-llm-comprehensive
description: Module demonstrating all LLM step fields (messages, tools, embeddings, multimodal)
tags: example, llm, comprehensive

params:
  - name: target
    required: true
  - name: scan_results
    default: ""
  - name: api_key
    generator: getEnvVar("OPENAI_API_KEY")

steps:
  # Basic chat completion with system/user/assistant messages
  - name: basic-chat
    type: llm
    messages:
      - role: system
        content: "You are a security analyst specializing in vulnerability assessment."
      - role: user
        content: "Analyze the following scan results for {{target}}"
      - role: assistant
        content: "I'll analyze the security findings and provide recommendations."
      - role: user
        content: "{{scan_results}}"
    llm_config:
      provider: openai
      model: gpt-4
      max_tokens: 1000
      temperature: 0.7
    exports:
      analysis_output: "{{response.content}}"

  # Multimodal content (text + image_url)
  - name: multimodal-analysis
    type: llm
    messages:
      - role: user
        content:
          - type: text
            text: "Analyze this screenshot for security issues"
          - type: image_url
            image_url:
              url: "{{Output}}/screenshot.png"
              detail: high
    llm_config:
      model: gpt-4-vision-preview
      max_tokens: 500
    exports:
      screenshot_analysis: "{{response.content}}"

  # LLM with tool definitions
  - name: llm-with-tools
    type: llm
    messages:
      - role: system
        content: "You are a security tool assistant. Use the provided tools to analyze targets."
      - role: user
        content: "Run a security scan on {{target}}"
    tools:
      - type: function
        function:
          name: run_nmap
          description: "Run an Nmap port scan on a target"
          parameters:
            type: object
            properties:
              target:
                type: string
                description: "The target IP or hostname to scan"
              ports:
                type: string
                description: "Port range to scan (e.g., '1-1000', '22,80,443')"
              scan_type:
                type: string
                enum: ["syn", "connect", "udp"]
                description: "Type of scan to perform"
            required:
              - target
      - type: function
        function:
          name: run_nuclei
          description: "Run Nuclei vulnerability scanner"
          parameters:
            type: object
            properties:
              target:
                type: string
                description: "Target URL to scan"
              templates:
                type: array
                items:
                  type: string
                description: "List of template categories to use"
            required:
              - target
    tool_choice: auto
    llm_config:
      model: gpt-4-turbo
      max_tokens: 2000
    exports:
      tool_calls: "{{response.tool_calls}}"

  # Tool choice: specific function
  - name: llm-specific-tool
    type: llm
    messages:
      - role: user
        content: "Scan {{target}} for open ports"
    tools:
      - type: function
        function:
          name: port_scan
          description: "Scan ports on a target"
          parameters:
            type: object
            properties:
              host:
                type: string
            required:
              - host
    tool_choice:
      type: function
      function:
        name: port_scan
    llm_config:
      model: gpt-4
    exports:
      forced_tool_call: "{{response.tool_calls}}"

  # Embedding generation
  - name: generate-embeddings
    type: llm
    is_embedding: true
    embedding_input:
      - "Security vulnerability found in {{target}}"
      - "SQL injection detected"
      - "XSS vulnerability present"
    llm_config:
      model: text-embedding-3-small
    exports:
      embeddings: "{{response.embeddings}}"

  # Response format for structured output
  - name: structured-output
    type: llm
    messages:
      - role: system
        content: "You are a security analyst. Output findings in JSON format."
      - role: user
        content: "List vulnerabilities found for {{target}}"
    llm_config:
      model: gpt-4-turbo
      max_tokens: 1000
      temperature: 0.3
      response_format:
        type: json_object
    exports:
      structured_findings: "{{response.content}}"

  # All llm_config fields
  - name: full-llm-config
    type: llm
    messages:
      - role: user
        content: "Provide a security assessment summary"
    llm_config:
      provider: openai
      model: gpt-4
      max_tokens: 500
      temperature: 0.5
      top_p: 0.9
      n: 1
      timeout: "60s"
      max_retries: 3
      stream: false
      custom_headers:
        X-Custom-Header: "security-scan"
    extra_llm_parameters:
      seed: 42
      presence_penalty: 0.1
      frequency_penalty: 0.1
    exports:
      summary: "{{response.content}}"
      model_used: "{{response.model}}"
      tokens_used: "{{response.usage.total_tokens}}"