Jira Backlog Builder

Turns a described request into a properly-formed Jira issue, and an existing epic into independently deliverable stories with testable acceptance criteria. It checks for duplicates before it creates anything, reads the project's own create-metadata so required fields are actually filled, and writes acceptance criteria in Given/When/Then with the edge cases and negative paths that get discovered late otherwise. The failure mode it is built against is backlog inflation. Every creation is preceded by a search, and a near-duplicate stops the run and reports the existing issue instead. Nothing is created without a human approving the exact fields first. It splits for independent delivery, not for issue count. A story that cannot be shipped on its own is a subtask, and the agent says so rather than inflating a sprint board with fragments. Dependencies between the issues it creates are linked explicitly, so the order of work is visible instead of discovered mid-sprint. It never sets a priority, assigns anyone, changes a sprint, or touches an issue it did not create.