Plan

---
name: plan
description: Maximum intelligence planning - dynamic agent selection, x--orchestrator orchestration, full Wisdom Council deliberation on every request. No shortcuts.
argument-hint: [feature or idea to explore]
---

# /plan - Maximum Intelligence Planning Protocol

You are initiating a planning session with **maximum intelligence**. Every request gets:
- **External web research** to find best practices, tools, and pitfalls
- Dynamic agent selection based on request type
- x--orchestrator orchestration
- Full 3-round Wisdom Council deliberation
- Confidence-calibrated recommendations

**No shortcuts. No quick mode. Maximum intelligence always.**

## Phase Overview

| Phase | Name | What Happens |
|-------|------|--------------|
| **0** | **Check Previous Retros** | **Read retros/ folder for learnings from past Ralph runs** |
| 1 | Understand & Clarify | Ask questions, explain terms, build context |
| 2 | Request Classification | Auto-classify type, announce agents with reasoning |
| 3 | **External Research** | **Web search for best practices, tools, pitfalls** |
| 4 | x--orchestrator Orchestration | Coordinate specialists, research codebase |
| 5 | Wisdom Council Deliberation | 3-round cross-examination, synthesis |
| **5.5** | **Project Alignment Validation** | **x--project-lead validates Council recommendation against PROJECT.md, cross-feature rules** |
| 5.6 | Design Validation | imp--ui-designer validates UX direction (when UI involved) |
| 6 | Present Findings | Show agent contributions, sources, agreement/disagreement |
| 7 | User Flow Narrative | Screenplay of the feature from user's perspective |
| 8 | Iterate Until Aligned | Loop with user until both agree |
| **8.5** | **Design Council (Optional)** | **Choose: "Ship it" or see 2-3 HTML mockup options before building** |
| 9 | Readiness Check & Handoff | Create alignment.md, hand off to /ralph |

---

## Phase 0: Check Previous Retrospectives (MANDATORY)

**BEFORE any planning begins, check for retros in `retros/` folder.**

### Step 1: Search for Retros

Use Glob tool: retros/RETRO-*.md Sort results descending (newest first by filename)

### Step 2: If Retros Exist

**2a. Read Most Recent Retro (PRIMARY)**
1. Take first file from sorted list (most recent)
2. Read the entire file
3. Find the `## 🔁 Execution Context` section - **CRITICAL for cross-session context**
4. Find the `## 🔴 FOR NEXT PLAN` section
5. Extract:
   - **Execution Context** → Key decisions, workarounds, codebase observations, project state
   - **Unresolved Issues** → These become T-FIX tasks in Layer 0
   - **Incomplete/Skipped Tasks** → Re-add to the new plan
   - **Patterns to Avoid** → Note these when planning
   - **Technical Debt** → Consider addressing

**2b. Scan Recent History (SECONDARY - Optional but recommended)**
1. Read the last 3-5 retros
2. Look for **recurring patterns**:
   - Same issues appearing multiple times?
   - Same task types failing repeatedly?
   - Patterns that worked consistently?
3. If recurring issues found, flag them as **systemic** (higher priority)

**2c. Tell the user what you found:**

Previous Run Context

Most Recent Retro: RETRO-2026-02-04-143022-feature-auth.md

• Status: [COMPLETE/PARTIAL/BLOCKED]
• Unresolved issues: [count]

Retro History: [X] total retros found

Execution Context from Terminal Session:

• Key decisions: [Summary of important decisions Ralph made]
• Workarounds discovered: [Solutions that aren't obvious]
• Project state: [What's working, what needs attention]

Unresolved Issues (will become T-FIX tasks):

• [Issue 1] ← from most recent
• [Issue 2] ← from most recent
• [Issue 3] ⚠️ RECURRING (appeared in 3 retros)

Incomplete Tasks (will be re-added):

• [Task 1]

Patterns to Avoid (from history):

• [Anti-pattern 1] - failed in last 2 runs

I'll incorporate these into the new plan.

### Step 3: If NO Retros Exist
Note: "No previous retrospectives found in retros/. Starting fresh."
Proceed to Phase 1.

### Phase 0 Completion

Phase 0 Complete: Previous Retrospective Check

What happened: Found X retros, Y unresolved issues / No previous retros found

What's next: Phase 1 - Understand and Clarify

---

## CRITICAL RULES

1. **ALWAYS ask clarifying questions** - Never assume you understand
2. **ALWAYS explain technical terms** - If you mention a tool/library/pattern, explain it in plain language
3. **ALWAYS classify and announce** - Tell user which agents you're engaging and WHY
4. **ALWAYS use x--orchestrator** - Delegate orchestration, don't do it yourself
5. **ALWAYS run Wisdom Council** - Full 3-round deliberation on EVERY request (no shortcuts)
6. **ALWAYS validate with x--project-lead AFTER Wisdom Council** - Project lead reviews the Council's recommendation against project-wide constraints. This prevents biasing the creative solution space while still catching conflicts before design work begins.
7. **ALWAYS show transparency** - Agent contributions, agreement/disagreement, confidence levels
8. **ALWAYS guide the user** - After each phase, explain what happened and what's next
9. **ONLY CREATE alignment.md** - /plan creates ONE file: `alignment.md` in project root. This hands off to /ralph. No other files (PRD.md, plan.md, activity.md, PROMPT_build.md are /ralph's job).
10. **ALWAYS do external research** - Search the web for best practices, tools, and pitfalls (unless trivially simple like CSS tweaks)
11. **ALWAYS be critical of sources** - Don't blindly trust what you find online; evaluate recency, credibility, and context fit
12. **ALWAYS cite sources** - Every recommendation must reference a URL, file:line, or prior session. If no source, explicitly label as "my reasoning."
13. **ALWAYS surface assumptions** - Before acting on any assumption, state it and ask the user to confirm. Example: "I'm assuming you want X because Y - is that right?"

---

## 📋 TodoWrite Phase Tracking (MANDATORY)

**IMMEDIATELY** when /plan starts, create a to