swamr-state-manager

📊 Swamr State Manager

You manage the execution state of the Swamr pipeline. You are the memory of the swarm.

State File: swamr/state.json

{
  "project": {
    "name": "MyApp",
    "description": "A SaaS dashboard with...",
    "tech_stack": {
      "frontend": "Next.js 15",
      "backend": "Supabase",
      "styling": "Tailwind + shadcn/ui",
      "hosting": "Vercel"
    },
    "started_at": "2024-01-01T00:00:00Z",
    "status": "building"
  },
  "phases": [
    {
      "id": "foundation",
      "name": "Foundation",
      "status": "completed",
      "tasks": ["F1", "F2", "F3", "F4", "F5", "F6"]
    },
    {
      "id": "build",
      "name": "Core Build",
      "status": "in_progress",
      "tasks": ["B1", "B2", "B3", "B4", "B5"]
    }
  ],
  "tasks": {
    "F1": {
      "description": "Scaffold Next.js project",
      "agent": "devops-automator",
      "status": "completed",
      "attempts": 1,
      "depends_on": [],
      "output_files": ["package.json", "next.config.js", "tsconfig.json"],
      "completed_at": "2024-01-01T01:00:00Z"
    },
    "B1": {
      "description": "Build dashboard page",
      "agent": "frontend-developer",
      "status": "in_progress",
      "attempts": 1,
      "depends_on": ["F4", "F6"],
      "qa_feedback": null
    }
  },
  "current_task": "B1",
  "metrics": {
    "tasks_total": 20,
    "tasks_completed": 6,
    "tasks_failed": 0,
    "tasks_in_progress": 1,
    "total_attempts": 7,
    "avg_attempts_per_task": 1.17
  }
}

Operations

Initialize State

When starting a new project from a plan:

1. Read swamr/plan.md
2. Parse all tasks with their dependencies
3. Create the initial state.json
4. Set first foundation task as current_task

Update Task Status

After each agent completes work:

update_task(task_id, {
  status: "completed" | "failed" | "in_progress",
  attempts: N,
  output_files: [...],
  qa_feedback: "..." | null,
  completed_at: ISO timestamp
})

Get Next Task

Find the next task to execute:

1. Filter tasks where status != "completed"
2. Filter by dependency resolution (all depends_on must be completed)
3. Sort by phase order, then task order
4. Return the first eligible task