Task Queue

Agent should:

1. Read tasks/queue.md
2. Present queued tasks to user, sorted by priority
3. User approves which tasks to work on
4. Move approved task to tasks/current.md, remove from queue.md

During /sprint

• After completing each task, check tasks/queue.md for new entries
• P0 tasks interrupt current work (present to user immediately)
• P1-P3 tasks are picked up in priority order after current task

Writing to the Queue (for external systems)

From Telegram Bot

# Append to tasks/queue.md in the project directory
with open(f"{project_path}/tasks/queue.md", "a") as f:
    f.write(f"\n## Queued: {datetime.now().isoformat()}\n")
    f.write(f"- **Source**: telegram\n")
    f.write(f"- **Task**: {task_description}\n")
    f.write(f"- **Issue**: {issue_id}\n")
    f.write(f"- **Priority**: {priority}\n")
    f.write(f"- **Agent**: auto\n\n")

From GitHub Action

- name: Queue task for Claude Code
  run: |
    echo "" >> tasks/queue.md
    echo "## Queued: $(date -Iseconds)" >> tasks/queue.md
    echo "- **Source**: github" >> tasks/queue.md
    echo "- **Task**: Deploy failed. Diagnose: ${{ steps.deploy.outputs.error }}" >> tasks/queue.md
    echo "- **Priority**: P0" >> tasks/queue.md
    echo "- **Agent**: devops" >> tasks/queue.md

From Cron/Event Bus

# Append to queue when project stalls
echo "" >> "$PROJECT_PATH/tasks/queue.md"
echo "## Queued: $(date -Iseconds)" >> "$PROJECT_PATH/tasks/queue.md"
echo "- **Source**: event-bus" >> "$PROJECT_PATH/tasks/queue.md"
echo "- **Task**: Project stalled >10 days. Run scan, identify blockers." >> "$PROJECT_PATH/tasks/queue.md"
echo "- **Priority**: P2" >> "$PROJECT_PATH/tasks/queue.md"
echo "- **Agent**: researcher" >> "$PROJECT_PATH/tasks/queue.md"

Queue Lifecycle

1. External system writes to tasks/queue.md
2. Session start notifies user of pending tasks
3. User approves task → moved to tasks/current.md
4. Task completed → removed from queue, logged in session
5. Empty queue.md after all tasks processed (or keep as log)

Priority Guide

• P0: Production down, data loss, security breach → interrupt everything
• P1: Feature broken for users, deadline approaching → next task
• P2: Normal work, improvements, non-urgent bugs → pick up in order
• P3: Nice-to-have, exploration, low-impact → when nothing else is queued