amonite.plan

Prompt file imported from thatbagu/3ds-type (.claude/commands/amonite.plan.md). Fill in {{arguments}} before use. Copyright stays with the author.

Create or update .amonite/plan.md from .amonite/spec.md, then apply the meta-environment section to the project flake.nix, and write the task dependency graph to .amonite/task-graph.json.

Gate: if spec.md has any unresolved Open questions, STOP and surface them. Do not proceed until the spec is clean.

If plan.md already exists, read it alongside tasks.md and clusters.nix before starting. You are extending the plan, not replacing it. Preserve existing cluster IDs (C001…) and task IDs (T001…); new clusters and tasks continue the sequence. Do not re-plan work that is already scaffolded (has a tasks/TNNN/task.nix file) or already verified (checked box in tasks.md).

1. Fix the technical context: language, dependencies, storage, target.
2. Design the architecture; name components — tasks will map onto them.
3. Write the verification strategy table. For every layer decide: hermetic (unit in task derivations, integration in cluster verify / nixosTest) or gate.live (impure, manual). Anything you cannot verify hermetically MUST appear in gate.live — hiding impurity violates principle E2.
4. Sketch the cluster topology in plan.md: which clusters exist, which tasks feed them, APP at the top. Write this to plan.md only — do NOT touch clusters.nix here. clusters.nix is written by /amonite.tasks once the task capsules exist. Updating it here, before task.nix files exist, will break nix flake check.
5. Apply the "Meta environment" package list to the project flake.nix devShell. Keep it minimal (principle E3): project-wide tools only; task-specific tools go into task envs later.
   • Find the devShells.*.default mkShell block.
   • If a # amonite:toolchain comment is already present, add new packages after it.
   • If the comment is absent, add it above the packages list:

     # amonite:toolchain — keep minimal; task tools go in task.nix env
     packages = [ ... ];

6. Write .amonite/task-graph.json — the machine-readable parallel execution plan. Include ALL tasks (existing + new). Format:

   {
     "waves": [
       {
         "wave": 1,
         "tasks": [
           { "id": "T001", "title": "...", "cluster": "C001", "depends": [] }
         ]
       },
       {
         "wave": 2,
         "tasks": [
           { "id": "T003", "title": "...", "cluster": "C001", "depends": ["T001"] }
         ]
       }
     ]
   }

   Rules: wave 1 = tasks with no dependencies; wave N = tasks whose depends are all in waves < N. Do not invent dependencies — if two tasks are truly independent, put them in the same wave.
7. Run nix flake check --no-build on the project root; the flake must evaluate cleanly. If it fails, fix the evaluation error before continuing — but do NOT create task.nix stubs to satisfy it. If the check fails because clusters.nix references tasks that don't exist yet, that means clusters.nix was edited incorrectly in this step; revert those changes and leave clusters.nix for /amonite.tasks.
8. Re-check the whole plan against .amonite/principles.md.

User input: {{arguments}}