AI_SCRUM/process/01_backlog_refinement.md

Process: Backlog Refinement

Primary agent: .claude/agents/product-owner.md (supported by dev-team for sizing)

When to run

Continuously, and especially before Sprint Planning, to keep 2-3 sprints’ worth of backlog items “ready.”

Steps

1. Pull the next unrefined items from the top of backlog/backlog.md / the relevant epic file in backlog/epics/.
2. Write/refine each as a user story with the Card/Conversation/Confirmation model (templates/user_story_template.md).
3. Attach explicit acceptance criteria to each story.
4. Have the dev-team agent provide a rough relative size (story points or T-shirt size) for each item.
5. Split any story too large to fit in a single sprint.
6. Check each item against templates/definition_of_ready.md; only mark “ready” if it passes.
7. Re-order backlog/backlog.md by value, risk, and dependency; document the reasoning briefly in its Notes section.

Exit criteria

• Top of backlog has enough “ready” items to fill the next 1-2 sprints.
• No ready item is missing acceptance criteria or a size estimate.

On completion: update state.md — phase → 2 (Sprint planning), leading agent → scrum-master, next action → run sprint planning for the next sprint. Refinement is continuous, so this phase can be re-entered any time the backlog runs low — it doesn't have to wait for a full sprint cycle.