plan

name	plan
description	Plan a .NET feature before coding. Spawns planning-orchestrator which researches, decomposes tasks, and writes `.claude/plans/{slug}/plan.md`. Use for any non-trivial change.
effort	high
argument-hint	<feature description> | --existing <plan.md>

/dotnet:plan

Produces a reviewable plan file at .claude/plans/{slug}/plan.md before any code is written.

When to Use

Signal	Use /dotnet:plan?
Feature touches >3 files	Yes
New entity / API / UI surface	Yes
Refactor crossing module boundaries	Yes
Unclear scope	Yes (scope classification is the first output)
Single-line fix	No — use /dotnet:quick
Bug debugging	No — use /dotnet:investigate
Pure question ("how does X work")	No — answer directly

Iron Laws (planning)

1. No implementation during planning — plan is about decisions, not code
2. Cite code with file:line — vague plans fail at execution
3. Every task is imperative — "Add validator" not "Should validate"
4. Classify scope — Tiny / Small / Medium / Large drives orchestration depth
5. Surface Iron Law implications at plan time — don't defer to review

Execution Flow

1. Delegate to planning-orchestrator (opus)
   • Subagent reads request, scans codebase
   • Spawns domain specialists in parallel (ef-schema-designer, api-architect, blazor-architect, etc. — only those relevant)
   • Runs context-supervisor to compress research
   • Drafts plan.md
2. For Large scope: orchestrator spawns Decision Council (dotnet-reviewer
   • one or two specialists) to critique the plan before you see it
3. Return: plan path, scope classification, task count, next command

Variants

Enhance existing plan

/dotnet:plan --existing .claude/plans/{slug}/plan.md

Research gaps, add missing tasks, flag risks. Doesn't rewrite — annotates.

Plan from triage

/dotnet:triage turns a review verdict into ready-to-work plans. Call plan to expand/refine a single triaged item.

Handoff

After plan written:

• /dotnet:brief .claude/plans/{slug}/plan.md — walk through the plan interactively (optional)
• /dotnet:work .claude/plans/{slug}/plan.md — start execution
• /dotnet:full — full lifecycle (plan + work + review + compound)

References

• ${CLAUDE_SKILL_DIR}/references/planning-workflow.md — end-to-end flow
• ${CLAUDE_SKILL_DIR}/references/plan-template.md — the plan.md structure
• ${CLAUDE_SKILL_DIR}/references/complexity-detail.md — Tiny/Small/ Medium/Large heuristics
• ${CLAUDE_SKILL_DIR}/references/agent-selection.md — which research agents to spawn for which signals
• ${CLAUDE_SKILL_DIR}/references/breadboarding.md — rough-sketch architecture before detail

Anti-patterns

• Planning the obvious — don't plan a 5-line fix
• Research theater — spawning 5 agents when 1 is enough
• "Investigate X" as a task — decide in research phase
• Copying the plan template without thinking — each section must have real content