Milknado Plan

Milknado Plan

Use this skill to turn a high-level goal into a structured Mikado execution plan backed by the Milknado graph. The skill calls the Milknado MCP server to inspect the current graph state, decompose the goal into fine-grained changes, and persist each change as a graph node.

When to use

• Starting a new feature, refactor, or bug fix that touches multiple files.
• Resuming an interrupted plan (add delta nodes without recreating existing ones).
• Validating that a set of proposed changes can be batched within a token budget.

Steps

1. Inspect the graph — call milknado_graph_summary to check for existing nodes. If nodes exist, operate in resuming mode.
2. Read context — read the relevant source files, spec documents, or issue descriptions needed to understand the goal.
3. Decompose — produce a v2 change manifest: file-level changes with id, path, edit_kind, description, symbols, and depends_on.
4. Validate batches — call milknado_plan_batches with the manifest to confirm the solver produces a feasible batch plan.
5. Persist nodes — call milknado_add_node for each change, linking prerequisite relationships via parent_id.
6. Summarise — report the number of nodes created, batch count, and any oversized or infeasible batches that need attention.

Milknado MCP tools

Manifest v2 quick reference

{
  "manifest_version": "milknado.plan.v2",
  "goal": "Short goal statement",
  "goal_summary": "What / why / success criteria (2–4 sentences).",
  "changes": [
    {
      "id": "c1",
      "path": "src/example.py",
      "edit_kind": "modify",
      "description": "Causal description referencing the goal.",
      "symbols": [{"name": "ExampleClass", "file": "src/example.py"}],
      "depends_on": []
    }
  ]
}

Closed edit_kind values: add, modify, delete, rename