plan

name	plan
description	Eight-lens planning with persona protocol. Writes to tasks/todo.md. STOP if sideways. Reviews gotchas first. Auto-loads skills. For 2+ step tasks.
allowed-tools	Read, Glob, Grep

Step 1: Context

Review tasks/gotchas.md + MEMORY.md + relevant docs.

Step 2: Run All 8 Personas

For the task at hand, answer each persona's questions:

• System Architect — fits system? data flow? boundaries? scale 10x/100x?
• Software Engineer — clean? SOLID? error handling? edge cases? types?
• Database Engineer — schema? indexes? RLS? queries? migrations reversible?
• QA Engineer — what breaks? null? empty? 10k? unicode? concurrent? auth bypass?
• Designer — production-grade? researched 3-5 real examples? no AI slop?
• Creative Director — visual identity? typography hierarchy? color story? spacing?
• UX Designer — frictionless? cognitive load? feedback? error recovery? progressive disclosure?
• Frontend Developer — RSC? accessible? mobile-first? all states? performance?

Document conflicts between personas and how they're resolved.

Step 3: Auto-Load Skills

Check CLAUDE.md skill matrix. Load ALL applicable skills for this task type. If UI is involved, load ALL design skills — no exceptions.

Step 4: Write Plan

Write plan to tasks/todo.md with:

• Checkable items per implementation step
• Persona checklist results (from Step 2)
• Skills loaded (from Step 3)
• Verification steps (6-phase loop)
• Estimate: S/M/L

STOP — wait for approval.

Step 5: Execute (after approval)

Backend → MiHCM components (npx @yashiel/mihcm-cli add) → Server → Client → Tests. If sideways: STOP and re-plan.

Step 6: Close

• Document results in tasks/todo.md
• Capture lessons → tasks/gotchas.md
• Update MEMORY.md with session context