planning-methodology

Planning Methodology Skill

Systematic approach for creating minimal, reversible implementation plans after research phase.

Auto-Invoke Triggers

• After ResearchPack is validated (score >= 80)
• Before code implementation begins
• When task requires multi-file changes
• User requests implementation plan

5-Step Methodology

Step 1: ResearchPack Review (30 sec)

• Verify ResearchPack score >= 80
• Extract required APIs and methods
• Note all gotchas and caveats
• List open questions

Step 2: Codebase Analysis (60 sec)

• Find existing patterns to follow
• Identify files to modify
• Check for similar implementations
• Map component dependencies

Step 3: Change Mapping (60 sec)

• List all files to modify/create
• Define minimal change set
• Identify dependencies between changes
• Plan test coverage

Step 4: Risk Assessment (30 sec)

• Identify potential failure points
• Assess impact on existing code
• Define rollback per change
• Flag breaking changes

Step 5: Plan Documentation (60 sec)

• Write structured ImplementationPlan
• Include verification per step
• Document rollback procedure
• Define success criteria

Output Format: ImplementationPlan

## ImplementationPlan: [Feature]

### Scope
- Files to modify: X
- New files: Y
- Tests to add: Z

### Changes

| # | File | Action | Verification |
|---|------|--------|--------------|
| 1 | path/file.ts | MODIFY | type-check |
| 2 | path/new.ts | CREATE | builds |

### Steps
1. [Step with verification]
2. [Step with verification]

### Rollback
[How to revert each change]

### Success Criteria
- [ ] npm run validate passes
- [ ] Feature works as specified

Planning Principles

Minimal Changes

• Only modify what's necessary
• No "while we're here" improvements
• Prefer small, focused changes

Reversibility

• Every change must be revertable
• Document rollback per step
• Keep backup strategy clear

Consistency

• Follow existing patterns
• Match naming conventions
• Use established architecture

Quality Gate

Note: the Plan >= 85 gate is owned by architect/analyst, which run on the opus tier (model: opus, effort: high) for higher reasoning rigor. See docs/model-tiers.md.

Plan must score >= 85:

Criterion	Points
All files listed	15
Step-by-step sequence	15
Verification per step	15
Rollback complete	15
APIs match research	15
Risk assessment	10
Test plan	10
Success criteria	5