Follow these steps in order:

Step 1 — Check project state:

python3 ~/.duplocloud/bin/duplo_ticket.py --check-project

• If exit code is 3: ask the user:

  "No active project found. Would you like to activate one now? (y/n)"

  • y → run the activate_project skill inline (follow its steps from the beginning), then re-run --check-project and continue.
  • n → stop.
• Capture from JSON output: project_name, has_execution_tasks, spec_empty, plan_empty.

Step 2 — Route based on execution tasks:

If has_execution_tasks is false:

Proceed directly to Step 3 with type spec_creation. Do not ask the user — they have already chosen to activate a ticket.

If has_execution_tasks is true:

Ask the user:

Execution tasks are ready for project <project_name>. Would you like to work on an execution task?

1. Yes — pick an execution task to work on
2. No — resume planning work (spec/plan)

Wait for user selection:

• If user picks 2: proceed to Step 3 with type spec_creation.
• If user picks 1: proceed to Step 2b.

Step 2b — Select an execution task:

python3 ~/.duplocloud/bin/duplo_ticket.py --list-execution-tasks

Parse the JSON output. Extract total and stages.

If total is 0: tell the user "No execution tasks found for this project." and stop.

If total > 10 — first ask the user to select a stage:

Show the numbered stage list:

Which stage would you like to work on?
1. <stage_name> (<N> tasks) — <stage_description>
2. <stage_name> (<N> tasks) — <stage_description>
...

Wait for the user's stage selection. Then show the tasks in that stage, including ticket info where available:

Which task would you like to work on?
1. <task_title> — <ticket_status> (<ticket_name>)
2. <task_title> — no ticket
...

If total <= 10 — show all tasks flat across all stages with stage context and ticket info:

Which task would you like to work on?
1. [<stage_name>] <task_title> — <ticket_status> (<ticket_name>)
2. [<stage_name>] <task_title> — no ticket
...

For each task: if ticket_name is present in the JSON, show <ticket_status> (<ticket_name>); otherwise show no ticket.

Wait for the user's task selection (number N). Resolve it to the definitive task name, title, and ticket info directly from the --list-execution-tasks JSON already in memory:

• If stage was selected (total > 10 flow): index into stages[S-1].tasks[N-1]
• If flat list (total <= 10 flow): flatten all stages[*].tasks into a single list and take item [N-1]

Capture name (UUID) as task_id and title as task_title. Use only these values going forward. Do not guess or recall UUIDs from earlier output.

Also capture ticket_name and ticket_id if present in the task object — these mean a ticket already exists for this task.

Step 2c — Check or create ticket for selected task:

• If ticket_name is present and non-empty in the --get-task-by-index output: the ticket already exists. Run:

  python3 ~/.duplocloud/bin/duplo_ticket.py --check-task-ticket --task-id <task_id>

  Then run Step 2d (inProgress), then Step 2e (load context), tell the user "Resuming existing ticket for task '<task_title>'. Here's what we've done so far: [brief summary from context]" and stop.

• If ticket_name is absent or empty: no ticket yet. Continue below.

List available agents:

python3 ~/.duplocloud/bin/duplo_ticket.py --list-agents

Show the numbered agent list to the user and ask:

Which agent should handle this task? (enter the number)

Wait for the user's selection. Look up the corresponding agent ID (id: value in parentheses). Then create the ticket:

python3 ~/.duplocloud/bin/duplo_ticket.py --create-execution-task-ticket \
  --task-id <task_id> \
  --agent-id <selected_agent_id>

• If exit code is 0: run Step 2d (inProgress), then Step 2e (load context), tell the user "Ticket created for task '<task_title>'. Let's get to work!" and stop.
• If exit code is 1: show the error output to the user and stop.

Step 2d — Move ticket to inProgress:

python3 ~/.duplocloud/bin/duplo_ticket.py --get-ticket-status

• If the output is open: run:

  python3 ~/.duplocloud/bin/duplo_ticket.py --set-ticket-status --status inProgress

• If the output is already inProgress or any other status: skip.

Step 2e — Load past context:

python3 ~/.duplocloud/bin/duplo_ticket.py --get-ticket-context

Read the full output carefully. It contains the mirrored conversation history from previous sessions on this ticket — prior user messages, Claude responses, decisions made, and work done. Use this to restore your understanding of where the work left off before responding to the user. If the output says "No past messages found", proceed without prior context.

Step 3 — Activate or create spec ticket:

python3 ~/.duplocloud/bin/duplo_ticket.py --activate-ticket --type spec_creation

• If exit code is 0 (ticket found and activated): run Step 2d (inProgress), then Step 2e (load context), then continue to Step 5.
• If exit code is 1 (not found): continue to Step 4.

Step 4 — Create spec ticket:

python3 ~/.duplocloud/bin/duplo_ticket.py --list-agents

Show the numbered agent list to the user. Write out each numbered item clearly so the user can read the full list. Then ask:

Which agent should handle this ticket? (enter the number)

Wait for the user's selection. Look up the corresponding agent ID from the list output (the id: value in parentheses on the selected line). Then create the ticket:

python3 ~/.duplocloud/bin/duplo_ticket.py --create-ticket \
  --type spec_creation \
  --agent-id <selected_agent_id>

• If exit code is 0: run Step 2d (inProgress), then Step 2e (load context), show the ticket creation output to the user, then continue to Step 5.
• If exit code is 1: show the error output to the user and stop.

Step 5 — Prompt based on spec/plan content state:

Using spec_empty and plan_empty captured in Step 1:

Case 1 — Both spec_empty and plan_empty are true:

Ask the user:

Both spec and plan are empty. What would you like to do?

1. Start writing the spec
2. Skip spec and go directly to plan creation

Case 2 — spec_empty is false, plan_empty is true:

Ask the user:

A spec exists but the plan is empty. What would you like to do?

1. Resume writing the spec
2. Move to plan creation

Case 3 — Both spec_empty and plan_empty are false:

Ask the user:

Both spec and plan already have content. Resume work on the plan?

• If the user confirms: delegate to skills/write_plan/SKILL.md.
• If the user declines: stop.

Step 6 — Delegate to write skill:

Based on the user's choice in Step 5:

• "Start writing spec" or "Resume writing spec" → read and follow skills/write_spec/SKILL.md.
• "Skip spec / go to plan", "Move to plan creation", or "Resume plan" → read and follow skills/write_plan/SKILL.md.