{
  "description": "Launch one supervised job worker with structured report, acceptance checks, audit artifacts, and strict parent retry boundaries.",
  "promptSnippet": "Launch one supervised job worker with a prompt, name, and optional acceptance contract.",
  "promptGuidelines": [
    "Use job narrowly: only when exactly one isolated supervised worker is more useful than direct root-agent work.",
    "Do not use job for ordinary main-quest work the root agent can handle well directly, or when the root agent needs to learn/debug process details directly.",
    "Do not use job to fan out multiple workers; for fan-out call jobs_plan with a matrix and a promptTemplate.",
    "Provide a clear name, concrete prompt, expected deliverables, and acceptance criteria when possible.",
    "Report policy: {{reportPolicy}}."
  ]
}