Context Optimizer

Optimize AI context loading for efficient token usage and focused development sessions.

When to Use This Skill

Use this skill when:

Working with large codebases (>50 files)
Context window is approaching limits
User mentions "context", "token", "optimize"
Need to focus on specific subsystem
Starting a new development session

Core Workflow

Step 1: Analyze Current Context

Identify loaded files:

Recently read files
Referenced documentation
Active working files

Assess relevance:

Current task requirements
File dependencies
Historical usage patterns

Step 2: Context Scoring

Score files by relevance (1-10):

Factor	Weight	Description
Direct relevance	40%	Directly related to current task
Dependency chain	25%	Required by relevant files
Recent access	20%	Recently read or modified
Reference frequency	15%	Often referenced in codebase

Step 3: Optimization Strategies

Strategy A: Essential Only

Load only:
- Files directly being modified
- Critical type definitions
- Immediate dependencies
Token savings: 60-80%

Strategy B: Focused Context

Load:
- Working files + 1 level dependencies
- Relevant documentation
- Key configuration
Token savings: 40-60%

Strategy C: Summarized Context

Load:
- Full working files
- Summaries of related files
- Index of available resources
Token savings: 30-50%

Context Summary Format

File Summary Template

## [filename] Summary
**Purpose:** [one-line description]
**Key exports:** [list of main functions/classes]
**Dependencies:** [key imports]
**Size:** [lines] lines

### Key Sections
- [Section 1]: Lines X-Y - [description]
- [Section 2]: Lines X-Y - [description]

Project Context Template

# Project Context Summary

## Architecture
- Pattern: [MVVM/MVC/etc]
- Language: [language + version]
- Framework: [framework details]

## Key Files
| File | Purpose | Priority |
|------|---------|----------|
| file1.cs | Main entry | High |
| file2.cs | Core logic | High |
| file3.cs | Utilities | Medium |

## Current Focus
Working on: [current task]
Relevant files: [list]

Output Format

Context Analysis Report

## Context Optimization Analysis

### Current Context
- Files loaded: 25
- Estimated tokens: ~45,000
- Utilization: 75%

### Recommended Optimization

**Strategy:** Focused Context
**Expected savings:** 40%

#### Keep (High Priority)
- ViewModel.cs - direct modification
- Model.cs - type definitions
- Services/*.cs - active dependencies

#### Summarize (Medium Priority)
- Utils/*.cs - create summaries
- Helpers/*.cs - create summaries

#### Defer (Low Priority)
- Tests/*.cs - load on demand
- Docs/*.md - reference only

### Action
Apply optimization? [Yes/No]

Integration with dev-docs-writer

This skill works with the dev-docs-writer agent for optimal context management:

Document Priority Loading

Priority 1 (Always Load):
  - docs/CONTEXT.md      # Quick reference, architecture snapshot
  - docs/PROGRESS.md     # Current phase, active tasks

Priority 2 (Phase-Specific):
  - docs/phases/phase-N/SPEC.md     # Current phase details
  - docs/phases/phase-N/TASKS.md    # Phase tasks
  - src/[active-module]/*           # Active development files

Priority 3 (On-Demand):
  - docs/PRD.md          # Requirements reference
  - docs/TECH-SPEC.md    # Technical details
  - src/**/*             # Specific files as needed

Session Continuity

## Starting a new session:

1. Load: docs/CONTEXT.md
2. Check: Current phase from PROGRESS.md
3. Load: Phase-specific files (docs/phases/phase-N/)
4. Resume: Work from last checkpoint

Token Budget Guidelines

Session Type	Token Budget	Loading Strategy
Quick check	~2K	CONTEXT.md only
Standard dev	~10K	CONTEXT + PROGRESS + active files
Deep dive	~30K	All docs + relevant source
Full context	~50K+	Complete project load

Phase-Aware Context Loading

자동 Phase 감지

# settings.json
context-optimizer:
  auto_load_phase_docs: true
  token_budget:
    quick: 2000
    standard: 10000
    deep: 30000
    full: 50000

Phase 문서 로딩 전략

Phase 감지 흐름:
1. PROGRESS.md에서 현재 Phase 확인
2. 해당 Phase 디렉토리 로드
3. SPEC.md → 범위 및 요구사항
4. TASKS.md → 현재 작업 목록
5. CHECKLIST.md → 완료 체크

Phase별 컨텍스트 템플릿

# Phase {{N}} Context Summary

## 현재 상태
- Phase: {{PHASE_NAME}}
- 진행률: {{PROGRESS}}%
- 활성 Task: {{ACTIVE_TASKS}}

## 핵심 파일
{{PRIORITY_FILES}}

## 현재 작업
{{CURRENT_WORK}}

## 참조 문서
- [SPEC.md](docs/phases/phase-{{N}}/SPEC.md)
- [TASKS.md](docs/phases/phase-{{N}}/TASKS.md)

세션 복구 워크플로우

새 세션 시작 시:

┌─────────────────────┐
│  CONTEXT.md 로드    │◀─── 필수
└─────────┬───────────┘
          │
          ▼
┌─────────────────────┐
│ PROGRESS.md 확인    │◀─── Phase N 감지
└─────────┬───────────┘
          │
          ▼
┌─────────────────────┐
│ Phase N 문서 로드   │◀─── SPEC + TASKS
└─────────┬───────────┘
          │
          ▼
┌─────────────────────┐
│ 작업 재개           │
└─────────────────────┘

토큰 예산별 Phase 로딩

예산	로드 범위	토큰
Quick	CONTEXT.md + PROGRESS.md	~2K
Standard	+ 현재 Phase (SPEC, TASKS)	~10K
Deep	+ 인접 Phase + 소스 코드	~30K
Full	모든 Phase + 전체 문서	~50K+

Sprint 통합

Phase와 Sprint 동시 사용 시:

Context Loading Priority:
  1. CONTEXT.md
  2. PROGRESS.md
  3. 현재 Sprint (sprints/sprint-N/)
  4. 연결된 Phase (phases/phase-N/)
  5. 소스 코드

Best Practices

Start Lean: Load minimum required context
Expand as Needed: Add files when referenced
Summarize Utilities: Keep only interfaces for helpers
Cache Summaries: Reuse context summaries across sessions
Document Dependencies: Track what requires what
Use Phase Documents: Leverage doc-splitter phase structure
Update PROGRESS.md: Record session outcomes for continuity
Phase-First: 현재 Phase 문서 우선 로드
Token Budget: 세션 유형에 맞는 토큰 예산 설정