nirvana context window optimizer

🔮✨ nirvana-context-window-optimizer ✨🔮
Optimize context window consumption across all your AI tools

v1.0.0  |  8 Agents  |  7 Tasks  |  MIT  |  AIOS Compatible

PT-BR | EN | ES | AR | HI | ZH | FR

Overview

The Nirvana Context Window Optimizer (NCWO) is a multi-agent squad that audits and optimizes all AI configuration directories in your project to reduce context window consumption by 30-70%.

It works on a safe copy (

output/

Value Proposition

• 30-70% reduction in token waste across AI configurations
• 12 platforms supported in a single run
• Zero risk to the original project (all changes go to

  output/

  )
• Comprehensive report with before/after metrics and migration guide
• 5-phase pipeline with automatic 20-point verification

Architecture

  PHASE 0          PHASE 1             PHASE 2           PHASE 3            PHASE 4
 Detection       Investigation       Strategy          Execution          Validation

+-----------+    +-----------+      +------------+    +------------+     +-----------+
|  🔍       |    |  📊       |      |  ♟️        |    |  📝        |     |  ✅       |
|  Mycroft  |--->|  Watson   |----->|  Irene     |--->|  Hudson    |---->|  Adler    |
|  Scanner  |    |  Auditor  |      |  Strategist|    |  File Opt  |     |  Verifier |
+-----------+    +-----+-----+      +------------+    +------+-----+     +-----------+
                       |                                     |                 |
                       v                                     v                 |
                 +-----------+                         +------------+          |
                 |  🕵️       |                         |  🏗️        |          |
                 |  Lestrade |                         |  Moriarty  |          |
                 |  Detective|                         |  Struct Opt|          |
                 +-----------+                         +------------+          |
                                                                              |
                       +------------------------------------------------------+
                       v
                +-------------+
                |   REPORT    |
                | OPTIMIZATION|
                |  REPORT.md  |
                +-------------+

                  🔮 Holmes (Orchestrator) coordinates all phases

Agent Table — The Baker Street Irregulars

Quick Start

# Activate the squad
/SQUADS:ncwo:ncwo-orchestrator

# Full optimization
*optimize /path/to/project

# Dry run (analysis only, no modifications)
*optimize /path/to/project --dry-run

# Quick mode (detection + audit only)
*optimize-quick /path/to/project

Supported Platforms

.claude/

.codex/

.gemini/

rules/

.cursor/

rules/

.antigravity/

.aios-core/

.agents/

skills/

.windsurf/

rules/

.github/copilot/

.aider/

.continue/

.cline/

rules/

Workflow Details

Phase 0 — Detection

• Agent: Mycroft (Scanner)
• Input: Project path
• Output: Inventory of detected AI platforms
• Action: Scans all 12 known directories, counts files, calculates sizes

Phase 1 — Investigation

• Agents: Watson (Auditor) + Lestrade (Detective)
• Input: Platform inventory
• Output: Per-file token report + detected anti-patterns
• Action: Counts real tokens, categorizes waste, detects problematic patterns

Phase 2 — Strategy

• Agent: Irene (Strategist)
• Input: Audit report + anti-patterns
• Output: Prioritized optimization plan (impact vs. risk)
• Action: Creates strategy ordered by maximum gain with minimum risk

Phase 3 — Execution

• Agents: Hudson (File Optimizer) + Moriarty (Structure Optimizer)
• Input: Optimization plan
• Output: Optimized files in

  output/

• Action: Applies content and structure optimizations without touching the original

Phase 4 — Validation

• Agent: Adler (Verifier)
• Input: Original project + optimized

  output/

• Output: Final report + migration guide
• Action: Runs 20-point checklist, generates before/after metrics

Anti-Patterns Detected

Verification Checklist (20 points)

Agent Adler runs a 20-point checklist split into blocking and advisory checks:

Blocking (12 points) — Failure prevents completion

output/

Advisory (8 points) — Reported but non-blocking

Output Structure

output/optimized-project/
├── .claude/              (optimized)
│   ├── CLAUDE.md
│   ├── rules/
│   └── settings.json
├── .codex/               (optimized)
├── .gemini/              (optimized)
├── .cursor/              (optimized)
├── .antigravity/         (optimized)
├── .agents/              (optimized)
├── .aios-core/           (assessment)
├── reports/
│   ├── OPTIMIZATION-REPORT.md
│   ├── MIGRATION-GUIDE.md
│   ├── token-savings.json
│   └── before-after.diff

Requirements

License

MIT -- Copyright (c) Luiz Gustavo Vieira Rodrigues

🔮 Part of the Nirvana Squad Ecosystem 🔮
Optimize. Verify. Evolve.

Created by Luiz Gustavo Vieira Rodrigues