Backlog Extractor
Plugin: tsh-product-management
Mention: @tsh-product-management:backlog-extractor
Model: sonnet
Tools: Read, Grep, Glob
Preloaded skill: extracting-epics-and-stories
Turns approved business intent into structured backlog content that a stakeholder can read without technical knowledge. It is designed so that a large backlog fans out: one agent per epic.
Responsibilities
| Area | Coverage |
|---|---|
| Inputs | The delegation must name the source files to read and the exact scope it owns: which epic or epics. If the scope or the inputs are missing, it stops and reports what is absent instead of widening the scope. PDFs are read with Read and an explicit pages range beyond 10 pages |
| Procedure | Follows the preloaded extracting-epics-and-stories skill and applies its templates exactly |
| Epic boundary | Stories belonging to another worker's epic are not its, even when the material discusses them |
| Traceability | Every story carries a Source pointing back to the material it came from |
| Delivery contract | Every epic carries a complete Epic Delivery Contract: customer outcome, end-to-end boundary, explicit exclusions, demonstrable scenario or criteria, known dependency or required shared contract, and the collaboration/concurrency classification. Downstream review and roadmap planning read those fields, and a missing one becomes a finding against the epic |
| Boundaries | No write tools, no interactive tool and no Atlassian tools. It never creates Jira tasks. Acceptance criteria stay as business-friendly GIVEN / WHEN / THEN scenarios with no implementation jargon |
What it returns
Its final message is the return value, with no preamble:
- Intent Brief, only when asked: goal, in and out of scope, stakeholders, likely epics, baseline overlap, open questions
- Epics: per epic its ID, title, 2 to 3 sentence business description, success criteria and delivery contract
- Stories: per story its ID, title, "As a… I want… So that…", Source, GIVEN/WHEN/THEN criteria, technical notes only if discussed, and a priority suggestion
- Cross-Epic Dependencies
- Assumptions
- Out of Scope
- Open Questions
Used by
orchestrating-business-analysis delegates intent brief drafting, epic identification and story extraction to it, through its delegating-to-workers.md reference (one agent per epic, or 2 to 3 small epics per agent). The orchestrator then merges the results and writes extracted-tasks.md itself.