TokenSaver

A token cost optimization skill that helps you save 50-80% on AI token usage without sacrificing response quality.

When to Use

Use TokenSaver when:

• - You have long conversations that consume many tokens
• You want to reduce AI API costs
• You're working with technical discussions that accumulate context
• You notice token usage growing rapidly in long sessions

Core Capabilities

1. Smart Context Compression

Automatically compresses conversation history based on message importance.

How it works:

• - Recent messages (last 3-5) kept fully intact
• Older messages summarized based on importance score
• Code blocks and critical decisions never compressed

Savings: 50-70% reduction in context tokens

2. Semantic Cache

Caches responses to similar queries to avoid reprocessing.

How it works:

• - L1: Exact query match → 100% savings
• L2: Semantic similarity > 85% → 80% savings
• L3: Pattern match → 50% savings

3. Adaptive Optimization

Automatically adjusts compression based on token pressure.

Stages:

• - < 3K tokens: No compression
• 3-6K tokens: Light compression
• 6-10K tokens: Medium compression
• > 10K tokens: Heavy compression + suggest new chat

Natural Language Commands

When user asks about TokenSaver in natural language, interpret and execute:

Settings & Configuration

User says: "Configure TokenSaver" / "TokenSaver settings" / "Setup TokenSaver"
Action: Show current configuration and available options
CODEBLOCK0

User says: "Use aggressive mode" / "Maximize savings" / "Set to save mode"
Action: Execute /tokensave command
Response: "✅ TokenSaver switched to aggressive save mode. This provides maximum token savings (up to 80%) with slight quality trade-off."

User says: "Use balanced mode" / "Default settings" / "Set to balanced"
Action: Execute /tokenbalance command
Response: "✅ TokenSaver switched to balanced mode. Good savings (50-70%) with quality preserved."

User says: "Prioritize quality" / "Keep full context" / "Set to quality mode"
Action: Execute /tokenquality command
Response: "✅ TokenSaver switched to quality priority mode. Minimal compression, maximum context retention for precision-critical tasks."

User says: "Disable TokenSaver" / "Turn off token saving" / "Stop TokenSaver"
Action: Execute /tokenoff command
Response: "⏸️ TokenSaver temporarily disabled. Re-enable anytime with /tokens or by asking me to turn it back on."

Usage Reports & Analytics

User says: "Show me token usage report" / "TokenSaver statistics" / "Token report"
Action: Execute /tokenreport command
Response format:
CODEBLOCK1

User says: "TokenSaver status" / "How much am I saving?" / "Token status"
Action: Execute /tokens command
Response format:
CODEBLOCK2

User says: "Clear token cache" / "Reset cache" / "Clear TokenSaver cache"
Action: Execute /tokencache clear command
Response: "✅ TokenSaver cache cleared. Fresh responses will be generated for subsequent queries."

Explanation & Help

User says: "What is TokenSaver" / "How does token saving work?" / "Explain TokenSaver"
Response:
CODEBLOCK3

User says: "TokenSaver not working" / "Why no savings?" / "TokenSaver not saving"
Response:
CODEBLOCK4

Best Practices

User says: "TokenSaver best practices" / "How to use TokenSaver effectively?" / "TokenSaver tips"
Response:
CODEBLOCK5

Slash Commands

For direct command access:

/tokens

Show current status and statistics CODEBLOCK6

/tokensave

Enable aggressive save mode

• - Maximum compression
• Best for very long technical discussions
• Slight quality trade-off possible

/tokenbalance

Balanced mode (default)

• - Good savings with quality preserved
• Recommended for most use cases

/tokenquality

Quality priority mode

• - Minimal compression
• Maximum context retention
• Use when precision is critical

/tokenreport

Generate detailed usage report CODEBLOCK7

/tokencache clear

Clear all cached responses

/tokenoff

Temporarily disable optimization

Usage Examples

Example 1: Long coding session
CODEBLOCK8

Example 2: Repeated questions
CODEBLOCK9

Example 3: Topic switching
CODEBLOCK10

Safety Features

TokenSaver never compresses:

• - Code blocks (always kept intact)
• Error messages and stack traces
• User-marked important messages
• Messages with high cross-references

Quality Guard:

• - Auto-rollback if quality drops > 15%
• One-click restore to uncompressed version
• Snapshots for every compression

Configuration

Default configuration:
CODEBLOCK11

Expected Results

Conversation Type	Tokens Saved	Quality Impact
Technical discussion (50 rounds)	70%	Minimal
Code review

80% | None | | Casual chat | 75% | None | | Quick Q&A | 30-50% | None |

Limitations

• - Requires conversation to exceed 3K tokens before compression starts
• First-time queries cannot be cached
• Very short conversations (< 10 messages) see minimal benefit
• Code-heavy conversations benefit most from smart referencing

Related Skills

• - shieldclaw: For security scanning
• browservisible: For web browsing
• filereader: For reading local files