work

Work Next Queue Item

Pick up the highest-priority pending item and begin working on it.

Process

1. Read .claude/queue/active.json and .claude/queue/backlog.json

2. Check if there's already an active item:

   • If yes, report its status and ask if you should continue or abandon it
   • If no, proceed to pick up next item

3. Select the next item from backlog:

   • Highest priority (lowest number) first
   • If same priority, oldest (earliest created) first
   • Skip items with blocked_by arrays containing unresolved task IDs — check if each blocking task exists in completed/ directory. If all blockers are completed, the task is unblocked.

4. Move the item from backlog to active:

   • Remove from backlog.json
   • Add to active.json with status: "in_progress"
   • Update the "updated" timestamp

5. Load company context before working:

   • Read .claude/company.yaml for tech stack, differentiators, market context
   • Read root CLAUDE.md for workflow rules and team structure
   • Read .claude/learnings/codebase.md for accumulated patterns

6. Report what you're starting:

   • Work item ID, title, and type
   • Problem to solve
   • Acceptance criteria (or research questions)
   • Files hint (if any)

7. Execute based on task type:

---

Branch A: Implementation Tasks (type: "implementation")

Standard implementation work.

1. If a plan_reference exists, read the referenced plan file first — it contains the approved approach
2. Follow story sizing rules (single context window, 3-5 files max)
3. Implement the solution
4. Apply goal-backward verification before completion
5. Use the appropriate specialist agent via Task tool if needed

On completion:

• Run goal-backward verification checklist
• Move item from active.json to .claude/queue/completed/{id}.json
• Update status to "completed"
• Add result object with summary, files_changed, and verification
• Generate completion report in .claude/queue/reports/{id}.md
• Report completion and what was accomplished