Function toolWorkflowSection

packages/lang-core/src/parser/prompt.ts:384–414

()

}

function toolWorkflowSection(): string {
  return `## Data Workflow

When tools are available, follow this workflow:
1. FIRST: Call the most relevant tool to inspect the real data shape before generating code
2. Use Query() for READ operations (data that should stay live) — NEVER hardcode tool results as literal arrays or objects
3. Use Mutation() for WRITE operations (create, update, delete) — triggered by button clicks via Action([@Run(mutationRef)])
4. Use the real data from step 1 as condensed Query defaults (3-5 rows) so the UI renders immediately
5. Use @-prefixed builtins (@Count, @Filter, @Sort, @Sum) on Query results for KPIs and aggregations — the runtime evaluates these live on every refresh
6. Hardcoded arrays are ONLY for static display data (labels, options) where no tool exists

WRONG — you called a tool and got data back, but you inlined the results:
\`\`\`
openCount = 2
item1 = SomeComp("first item title")
item2 = SomeComp("second item title")
list = Stack([item1, item2])
chart = SomeChart(["A", "B"], [12, 8])
\`\`\`
This is static — it shows stale data and won't update. Creating item1, item2, item3... manually is ALWAYS wrong when a tool exists.

RIGHT — use Query() for live data, Mutation() for writes, @builtins to derive values:
\`\`\`
data = Query("tool_name", {}, {rows: []})
openCount = @Count(@Filter(data.rows, "field", "==", "value"))
list = @Each(data.rows, "item", SomeComp(item.title, item.field))
createResult = Mutation("create_tool", {title: $title})
submitBtn = Button("Create", Action([@Run(createResult), @Run(data), @Reset($title)]))
\`\`\`
Everything derives from the Query — when data refreshes, the entire dashboard updates automatically.`;
}

function importantRules(
  rootName: string,