backlog-worker

name	backlog-worker
version	1.0
description	Executes backlog items one at a time — selects the next ready item, decomposes it into atomic actions, performs each action using the appropriate sub-skill, reviews the output, records learnings, and advances the item to done. Use when asked to work the backlog, execute the next item, or make autonomous progress on defined work.

Skill: Backlog Worker

When Not to Use

• When the task is to add, refine, list, start, complete, or archive backlog items rather than execute the work they describe — use backlog-manager instead
• When no items are in ready status — apply backlog-manager to refine items to ready before invoking this skill
• When the item requires a human decision or external approval before work can proceed — stop and surface the decision rather than proceeding autonomously
• When the scope of an item cannot be completed in a single session and partial completion cannot be verified — record progress, leave the item active, and stop rather than marking it done

---

Interaction Protocol

Before starting, ask if not already clear:

1. Which item to work — a specific item ID, or should the skill select the first ready item automatically?
2. Are there existing learnings records or prior research relevant to this work?
3. What constitutes done for this item — what observable state must be true before the item can be marked complete?
4. Should the skill stop at the first blocker, or continue to the next atomic action where possible?