token-saver – OpenClaw Skill

---

name: token-saver version: 2.0.0 description: "Reduce OpenClaw AI costs by optimizing workspace files, model selection, and chat compaction. Analyzes .md files for compression, audits AI model usage, and recommends chat compaction thresholds. Writes to AGENTS.md for persistent mode. Accesses home directory to read OpenClaw config and session history." author: RubenAQuispe repository: https://github.com/RubenAQuispe/token-saver

Token Saver

💡 Did you know? Every time you send a prompt, your workspace files (SOUL.md, USER.md, MEMORY.md, AGENTS.md, and more) are sent along with it — every single time. These files count toward your context window, slowing down responses and costing you real money on every message.

Token Saver compresses these files using AI-efficient notation that preserves all your data while making everything lighter, faster, and cheaper.

Commands

Command	What it does
/optimize	Full audit dashboard — files, models, compaction
/optimize tokens	Compress workspace files (auto-backup)
/optimize compaction	Chat compaction control
/optimize compaction 120	Set custom compaction threshold
/optimize revert	Restore files from backups, disable persistent mode

Features

📁 Workspace File Compression

Scans all .md files sent with each API call. Shows token count and potential savings per file.

File-aware compression:

• SOUL.md — Light compression, keeps evocative/personality language
• AGENTS.md — Medium compression, dense instructions
• USER.md / MEMORY.md — Heavy compression, key:value data
• PROJECTS.md — No compression (user-specific structure)

🤖 Model Audit

Detects current AI models for main chat, cron jobs, and subagents. Suggests cheaper alternatives where appropriate (e.g., Gemini free tier for background tasks).

💬 Chat Compaction

Analyzes your conversation history (default: last 7 days) to recommend optimal compaction threshold.

How it works:

• Scans your sessions to detect topic change patterns
• Recommends threshold based on YOUR usage
• Shows savings estimate per preset

Presets:

• 🟢 Safe (160K) — ~$30/mo savings
• 🟡 Balanced (120K) — ~$100/mo savings
• 🔴 Aggressive (80K) — ~$200/mo savings

Options:

• --month — Scan last 30 days (more data)
• --all — Scan all history (takes longer)
• Custom number: /optimize compaction 100 = compact at 100K tokens

📝 Persistent Mode

When enabled, adds file-type-aware writing guidance to AGENTS.md:

File	Writing Style
SOUL.md	Evocative, personality-shaping
AGENTS.md	Dense instructions, symbols OK
USER.md	Key:value facts
MEMORY.md	Ultra-dense data
memory/*.md	Log format, dated
PROJECTS.md	Keep user's structure

Safety

• Auto-backup — All modified files get .backup extension
• Integrity > Size — Never sacrifices meaning for smaller tokens
• Revert anytime — /optimize revert restores all backups
• No external calls — All analysis runs locally

Installation

clawhub install token-saver

Version

2.0.0 — Chat compaction, file-aware compression, persistent mode

🚀 Token Saver

💡 Did you know? Every time you send a prompt, your workspace files (SOUL.md, USER.md, MEMORY.md, AGENTS.md, and more) are sent along with it — every single time. These files count toward your context window, slowing down responses and costing you real money on every message. Token Saver compresses these files using AI-efficient notation that preserves all your data while making everything lighter, faster, and cheaper.

Cut your OpenClaw AI costs by 40-90% with one command.

Quick Start

/optimize

That's it. You'll see a dashboard with your savings options.

What It Does

Feature	Description
🗜️ File Compression	Scans ALL .md workspace files and compresses them using AI-efficient notation
🤖 Model Audit	Detects your current AI models and suggests cheaper alternatives
📊 Cost Dashboard	Shows weekly/monthly/annual savings with clear before & after
🛡️ Safe Rollback	Auto-backups + one-command revert

Example

Before (500+ tokens):

When Ruben greets me in the morning with phrases like "good morning" or "what's on today", I should proactively review our task list, mention pending items, and check for urgent issues...

After (30 tokens):

MORNING: greeting → review(todos+pending+urgent)

Same meaning. 90% fewer tokens. Real dollar savings.

Commands

Command	What It Does
/optimize	Show savings dashboard
/optimize tokens	Compress workspace files (auto-backup)
/optimize models	Detailed model cost comparison
/optimize revert	Restore all files from backups

Dashboard Preview

🚀 Token Optimizer Dashboard

💾 Current Context: 12,169 tokens across 10 files
💰 Est. Monthly Cost: $158.08

┌─────────────────────────────────────────────┐
│  🗜️  WORKSPACE FILES OPTIMIZATION           │
└─────────────────────────────────────────────┘
🔴 MEMORY.md:  2,640 → 215 tokens (92% possible saving)
🔴 USER.md:      563 → 103 tokens (82% possible saving)
🟢 AGENTS.md:  2,063 → 2,063 tokens (0% possible saving)

┌─────────────────────────────────────────────┐
│  🤖  AI MODEL AUDIT                         │
└─────────────────────────────────────────────┘
• Default: claude-sonnet-4 (~$14.40/month)
• Cron jobs: gemini-pro (free)
💡 Subagents: Switch Opus → Sonnet — ~$14.40/month possible saving

┌─────────────────────────────────────────────┐
│  📊 COMBINED POSSIBLE SAVINGS               │
└─────────────────────────────────────────────┘
File compression: ~$37.40/month
Model switching:  ~$17.57/month
Total:            ~$54.97/month possible saving

Install

# From ClawHub
clawhub install token-saver

# Or clone directly
git clone https://github.com/RubenAQuispe/token-saver.git

✨ Persistent Mode — One-and-Done

When you run /optimize tokens, Token Saver also enables Persistent Mode — your AI will continue writing in compressed notation going forward. No need to re-optimize!

• Automatic — Enabled when you optimize, disabled when you revert
• Smart — AI maintains compressed format in all workspace files
• Reversible — /optimize revert turns it off and restores everything

Safety First

• ✅ Auto-backup before any file changes
• ✅ "Possible savings" shown until you actually apply
• ✅ One-command revert — /optimize revert restores everything + turns off persistent mode
• ✅ Only compresses files where real savings exist

How It Works

AI models understand compressed notation perfectly. The optimizer converts verbose instructions into dense, structured formats that preserve 100% of the meaning while using 40-90% fewer tokens.

Every token saved = money saved on every single API call.

License

MIT — Use it, modify it, share it.