Phase 1: Plan Discovery

1. Locate the plan to validate:

   • Check for plan file path in conversation context
   • Look in .claude/plans/ for recent plan files
   • If no plan file, check if plan was stated inline in conversation

2. Extract plan content and metadata:

   • Files to be modified
   • Steps/tasks count
   • Dependencies mentioned

3. If plan >50 lines, use AskUserQuestion to confirm scope

Phase 2: Plan Validation (5 Parallel)

Launch ALL 5 agents in a single message with multiple Task calls. See references/phase-2-plan-validators.md for agent prompts.

Provide each agent:

• Full plan content
• Original user request (from conversation context)
• Relevant pattern file content (load from claude-patterns/)

Phase 3: Devil's Advocate

Launch 1 agent to challenge ALL findings from Phase 2. See references/phase-3-devil-advocate.md for agent prompt.

Provide:

• All findings from Phase 2
• Plan content for context verification

Output: Validated findings with status (CONFIRMED/DOWNGRADED/DISMISSED/UPGRADED)

Phase 4: Report + User Decision

1. Generate report using templates/plan-assessment.md
2. Present findings grouped by severity (P0, P1, P2, P3)
3. Use AskUserQuestion:

{
  questions: [
    {
      question: "How would you like to proceed with this plan?",
      header: "Action",
      options: [
        {
          label: "Revise plan",
          description:
            "Update plan to address P0/P1 findings before implementing",
        },
        {
          label: "Proceed as-is",
          description: "Accept the plan and start implementation",
        },
        {
          label: "Start over",
          description: "Request a completely new plan approach",
        },
      ],
      multiSelect: false,
    },
  ];
}