🧠 MOS

Multiskill Optimum Saver

AI optimization layer for Claude Code — match processing power to task complexity, automatically.

MOS is an AI optimization layer designed to streamline work with Claude-based development tools through intelligent resource management. It solves the problem of token waste and unnecessary costs that occur when overpowered models are used for simple tasks — or conversely, when insufficient power degrades output quality on complex ones.

Using a scoring algorithm (0–140), the system weighs parameters such as model type, thinking depth, text compression, and sub-agent count. Users can choose from five predefined levels — from an economical Trivial tier to a powerful Expert tier — or make manual adjustments. Throughout the session, MOS monitors incoming requests and suggests real-time configuration updates to ensure optimal alignment between task complexity and allocated processing power.

The goal: maximize productivity and cut operational costs by 30–75% — without compromising output quality.

What It Does

At the start of every Claude session, MOS:

1. Reads your current session-config.json
2. Scans project context (file count, git history, CLAUDE.md)
3. Classifies complexity: TRIVIAL → SIMPLE → MEDIUM → HARD → EXPERT
4. Scores your config (0-140) against complexity requirements
5. Displays a status block — match, borderline, or mismatch
6. Recommends specific changes only when needed
7. Applies changes on your approval (yes / no)

Status Block (every session)

╔═ MOS ═════════════════════════════════════════╗
║  Model: sonnet        Sub: haiku×3            ║
║  Caveman: lite        Thinking: low           ║
║  Compact@: 60%        Score: 64/140           ║
╠═══════════════════════════════════════════════╣
║  Session: MEDIUM      Reason: refactor task   ║
║  Status:  ✓ Config matches session needs      ║
╚═══════════════════════════════════════════════╝

Presets by Complexity

| Complexity | Model | Caveman | Thinking | Subagents | |------------|--------|---------|----------|-----------| | TRIVIAL | haiku | ultra | off | 0 | | SIMPLE | sonnet | full | off | 1×haiku | | MEDIUM | sonnet | lite | low | 3×haiku | | HARD | sonnet | lite | high | 4×sonnet | | EXPERT | opus | off | high | 5×sonnet |

Scoring (0-140)

| Parameter | Points | |-----------|--------| | haiku | +18 | | sonnet | +58 | | opus | +100 | | thinking low/high/max | +8/+22/+38 | | caveman lite/full/ultra | -2/-8/-16 | | each haiku subagent | +4 (max 28) | | each sonnet subagent | +7 (max 28) |

Multi-Skill Routing (automatic)

| Trigger | Skill | |---------|-------| | z-index, CSS stuck, RTL, PNG | css-expert | | ad creative, image prompt, copy | ad-creative | | mobile, iOS, Android, responsive | mobile-inspector | | Word, DOCX, PDF | docx | | context > 60% | strategic-compact |

Installation

Easy install (recommended)

npx mos-claude

Hebrew version:

npx mos-claude --hebrew

That's it. One command — MOS is ready. Caveman skill is installed automatically as a required dependency.

Alternative: global install

npm i -g mos-claude
mos-install            # English
mos-install --hebrew   # Hebrew

Alternative: clone & run

git clone https://github.com/Yula-Digital/MOS.git
cd MULTISKILL_OPTIMUM_SAVER
bash install.sh           # English
bash install.sh --hebrew  # Hebrew

Windows: Right-click Install-MOS.ps1 → Run with PowerShell

claude.ai: Paste SKILL_EN.md into Project Instructions.

Restart Claude Code / Claude Desktop. Done.

Commands

| Command | Action | |---------|--------| | /mos | Show config status + level menu | | /mos [1-5] | Apply preset | | /mos reset | Restore defaults | | /mos save | Save current as default |

Companion Tools

• Caveman — 50-75% token reduction
• Codebase Memory MCP — 99% reduction
• Claude-Mem — cross-session memory
• Token Savior — -77% tokens

License

MIT — free to use, modify, and distribute.

⭐ Star if useful | 🐛 Issues | 💬 Discuss