Token Savings

Show how many tokens cc-soul has saved by avoiding redundant operations.

Execute directly — Runs the token-savings.sh script and shows report.

How It Works

The soul tracks three types of token savings:

1. Recall Hits

When you ask about something and the soul already knows it, we avoid re-reading files:

• File re-read avoided: ~500-5000 tokens per file
• Exploration avoided: ~2000-10000 tokens per search

2. Codemap Cache Hits

When the codemap is cached and git commit unchanged:

• Full codebase scan avoided: ~1000-5000 tokens
• File tree generation avoided: ~500-2000 tokens

3. Transparent Memory Injection

Context injected from memory instead of exploration:

• Pre-answered questions: ~500-2000 tokens per injection
• Pattern recall: ~200-1000 tokens per pattern

Usage

/token-savings          # Show report
/token-savings reset    # Reset counters

Implementation

When this skill is invoked:

${CLAUDE_PLUGIN_ROOT}/scripts/token-savings.sh report

Example Output

=== CC-Soul Token Savings Report ===

Recall (avoided file re-reads):
  Hits: 12
  Chars saved: 45000 (~11250 tokens)

Codemap cache (avoided re-scans):
  Hits: 3
  Chars saved: 8000 (~2000 tokens)

Transparent memory (injected context):
  Injections: 45
  Chars injected: 22500 (~5625 tokens)

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
TOTAL ESTIMATED SAVINGS: ~18875 tokens

Accuracy Notes

Token savings are estimates based on:

• 4 characters ≈ 1 token (rough average)
• Assumes alternative would be file read or exploration
• Actual savings vary by use case

The real benefit is qualitative: faster responses, less context churn, preserved working memory across sessions.