context-maximizer

Files (8)

• INSTALL.md
• QUICK_START.md
• README.md
• SKILL.md
• SUMMARY.md
• posts-crud-example.md · examples/
• loading-strategies.md · references/
• analyze_context.py · scripts/

SKILL.md (first lines)

# Context Maximizer

Leverage the 1M token context window (GA since March 2026, available on both Opus 4.6 and Sonnet 4.6) to provide full codebase awareness during spec generation, planning, and development. With 64k default output tokens (up to 128k on Opus 4.6) and support for up to 600 images/PDF pages per request, smart loading strategies ensure the most relevant context is always available.

## When to Use

- During spec generation (specify skill)
- When creating technical plans (plan skill)
- Before implementing features that require understanding multiple systems
- When analyzing dependencies across the codebase
- For comprehensive code reviews requiring full context

## Context Budget

Total available: **1,000,000 tokens** (~4,000,000 characters)

### Priority Levels

1. **Critical** (always loaded): Constitution, active specs, current plan
2. **High**: Skills relevant to feature domain
3. **Medium**: Codebase files matching feature patterns
4. **Low**: Documentation, references, examples

### Token Allocation Strategy

```
Constitution:       ~10,000 tokens   (1%)
Active Specs:       ~50,000 tokens   (5%)
Current Plan:       ~20,000 tokens   (2%)
Skills (3-5):      ~100,000 tokens  (10%)
Codebase:          ~700,000 tokens  (70%)
Documentation:      ~50,000 tokens   (5%)
Buffer:             ~70,000 tokens   (7%)
```

## Features

### 1. Context Budget Calculator

Analyzes project structure and estimates token usage:

```bash
python scripts/analyze_context.py /path/to/project
```

Output:
```json
{
  "total_tokens": 850000,
  "by_category": {
    "constitution": 8500,
    "specs": 42000,
    "skills": 95000,
    "codebase": 650000,
    "docs": 54500
  },
  "within_limit": true,
  "utilization": "85%"
}
```

### 2. Smart Loading Strategy

Automatically determines what to load based on feature domain:

```bash
python scripts/analyze_context.py /path/to/project --feature "posts-crud"
```

Output includes prioritized file list:
```json
{
  "loading_plan": {
    "critical": [
      "constitution.yml",
      "specs/1-posts-crud/prd.json",
      "specs/1-posts-crud/plan.json"
    ],
    "high": [
      "skills/speckit/specify/",
      "skills/speckit/plan/",
      "skills/database/"
    ],
    "medium": [
      "src/db/schema/posts.ts",
      "src/lib/services/posts.ts",
      "src/app/api/posts/"
    ],
    "low": [
      "docs/database.md",
      "docs/api-integrations.md"
    ]
  }
}
```

### 3. Context Report

Shows what's loaded and provides optimization suggestions:

```bash
python scripts/analyze_context.py /path/to/project --report
```

Output:
```json
{
  "loaded": {
    "files": 156,
    "total_tokens": 820000,
    "breakdown": {
      "constitution": { "tokens": 8500, "percentage": 1.0 },
      "specs": { "tokens": 42000, "percentage": 5.1 },
      "skills": { "tokens": 95000, "percentage": 11.6 },
      "codebase": { "tokens": 620000, "percentage": 75.6 },
      "docs": { "tokens": 54500, "percentage": 6.6 }
    }
  },
  "warnings": [],

From JimmyBlanquet/project-forge, MIT.

More from JimmyBlanquet/project-forge