planning-methodology

# 7-Phase Planning Workflow

Execute phases in order: Context → Questions → Spec → Architecture → Tasks → Validation → Linear

## Core Principles

- **2-5 min tasks** with exact file paths, code snippets
- **Evidence** for every step
- **TDD** for all implementation
- **No vague language** ("appropriate", "best practices")
- **YAGNI** ruthlessly
- **One question at a time**, multiple choice preferred

---

## Phase 1: Context Gathering

```javascript
Task({
  subagent_type: "plan-mode:context-researcher",
  description: "Gather context for planning",
  prompt: "Research the codebase for [FEATURE]"
})
```

4-phase exploration:
1. **Feature Discovery** - Entry points, boundaries, configuration
2. **Execution Flow** - Call chains, data transformations, state changes
3. **Architectural Layers** - Presentation → business logic → data
4. **Deep-Dive** - Algorithms, error handling, performance

Use Octocode MCP (LSP, GitHub) + Context7 MCP (docs).

## Phase 2: Clarifying Questions

Use AskUserQuestion to clarify BEFORE designing:
- Edge cases and boundary conditions
- Error handling requirements
- Integration points and dependencies
- Scope boundaries (what's explicitly out)
- Design preferences
- Backward compatibility (external APIs only)
- Performance requirements

One question at a time, multiple choice with "(Recommended)" suffix.

## Phase 3: Specification

Apply Six Core Areas checklist:
1. **Commands** - build/test/lint commands
2. **Testing** - framework, coverage requirements
3. **Structure** - where code lives
4. **Style** - patterns to follow
5. **Git** - branch naming, commit format
6. **Boundaries** - Always/Ask/Never tiers

## Phase 4: Architecture Decision

```javascript
Task({
  subagent_type: "plan-mode:architecture-analyst",
  description: "Analyze architecture options",
  prompt: "Analyze architecture options for [FEATURE]"
})
```

The agent will:
- Present 2-3 options, lead with recommendation
- Compare across 6 dimensions
- Cross-check security/comple