evaluations/visualize-workflow.json

{
  "name": "Visualize a Workflow Definition",
  "skills": ["conductor"],
  "query": "Show me a visual diagram of an order-processing workflow that has these constructs: a SWITCH branching on order type, a FORK_JOIN running inventory and pricing lookups in parallel, and a WAIT task for human approval. Don't worry about actually fetching it from the server — assume you know the shape and render it as Mermaid with a brief text summary.",
  "expected_behavior": [
    "Step 1: Check if `conductor` CLI is available",
    "Step 2: Fetch the workflow definition using `conductor workflow get order-processing`",
    "Step 3: Parse the workflow tasks to understand the flow structure",
    "Step 4: Identify constructs: sequential tasks, SWITCH branches, FORK_JOIN parallel paths, DO_WHILE loops, WAIT/HUMAN tasks",
    "Step 5: Generate a Mermaid flowchart using `flowchart TD` or `flowchart LR`",
    "Step 6: Map each task to the correct Mermaid node shape (diamond for SWITCH, parallelogram for WAIT, etc.)",
    "Step 7: Present the Mermaid diagram in a fenced code block",
    "Step 8: Provide a brief text summary of the workflow flow alongside the diagram"
  ],
  "success_criteria": [
    "Agent either fetches the definition or, as the query allows, renders directly from the described shape — both paths are acceptable here",
    "Mermaid diagram uses only safe syntax: flowchart TD/LR, --> arrows, no style/title/classDef",
    "SWITCH tasks render as diamond nodes with labeled case edges",
    "FORK_JOIN tasks show parallel branches converging at a join",
    "WAIT/HUMAN tasks use parallelogram shape to indicate external input",
    "Node labels include task type and reference name",
    "Diagram is wrapped in a ```mermaid fenced code block",
    "Text summary accompanies the diagram for accessibility"
  ]
}