token-efficiency-audit

name	token-efficiency-audit
description	Run a comprehensive token-efficiency audit on Claude usage patterns — identify input/output token waste, context-window inefficiency, low cache-hit rates, and surface 11 standard optimization patterns (O6a-O6e). Use when the user wants to audit token spend, profile prompt cost, or align with SOSA Level 3+ token-economy requirements.

Token Efficiency Audit Skill

Overview

This skill performs a comprehensive token efficiency audit on Claude usage patterns, identifying optimization opportunities and generating actionable recommendations aligned with SOSA Level 3+ requirements.

Metadata

• Skill ID: token-efficiency-audit
• Version: 1.0.0
• SOSA Level: 3 (Secured + Supervised)
• Audit Coverage: Input tokens, output tokens, context window efficiency, cache hit rates
• Optimization Patterns: 11 standard patterns (O6a-O6e categories, including O6c-4 Redundant Context Detection)

Execution Model: Plan → Act → Verify

Phase 1: Plan

Analyze audit scope and establish baseline metrics:

1. Define Audit Boundaries

   • Time period (last 30 days, last quarter, custom range)
   • User/team scope (individual, department, organization)
   • Model coverage (Claude 3 family, multimodal, specific versions)
   • Environment (production, staging, development)

2. Establish Baselines

   • Current token consumption (daily, weekly, monthly averages)
   • Cost per 1M tokens (input, output, cache read/write)
   • Context window utilization rate
   • Cache effectiveness metrics

3. Define Success Metrics

   • Target reduction percentage (typically 15-30%)
   • Cost ceiling per operation
   • Latency targets
   • Quality thresholds (accuracy/helpfulness maintenance)

Phase 2: Act

Execute optimization analysis against 10 standard patterns:

1. Pattern Categories

   • O6a: Prompt compression and summarization
   • O6b: Context window optimization
   • O6c: Batch processing and request consolidation
   • O6d: Cache strategy optimization
   • O6e: Model selection and routing

2. Analysis Steps

   • Extract top 20% highest-cost interactions (80/20 analysis)
   • Apply each pattern category to high-cost interactions
   • Calculate theoretical token savings
   • Identify implementation complexity and risk
   • Apply O6c-4 (Redundant Context Detection): check marketplace plugins against built-in features and user commands for duplicates; classify MCP connectors as filterable (.mcp.json) vs non-filterable (mcp__claude_ai_*)

3. Output Generation

   • Pattern match report (which patterns apply to which usage)
   • Per-interaction optimization scores
   • Implementation priority matrix
   • Risk assessment per recommendation

Phase 3: Verify

Validate recommendations and measure impact:

1. Validation Checks

   • Ensure recommended changes maintain output quality
   • Verify SOSA compliance (security, audit, supervision)
   • Confirm implementation feasibility
   • Cross-reference with compliance checklist

2. Impact Measurement

   • Projected token savings (conservative, realistic, optimistic)
   • Cost impact (absolute $ and % reduction)
   • Latency impact (positive/negative/neutral)
   • Implementation effort (hours, complexity)

3. Recommendations Ranking

   • Quick wins (high savings, low effort)
   • Strategic optimizations (medium savings, medium effort)
   • Long-term improvements (lower immediate impact, architectural)

Key Metrics

Token Efficiency Score

Score = (Baseline Tokens - Recommended Tokens) / Baseline Tokens × 100
Range: 0-100%
Target: 20-35% for mature systems

Implementation Priority

Priority = (Projected Savings × Feasibility) / (Effort × Risk)
Sort descending for execution order

Optimization Patterns Reference

See references/optimization-patterns.md for detailed pattern descriptions, implementation examples, and per-pattern token savings estimates.

Integration Points

• Supervised: All recommendations require human review before implementation
• Orchestrated: Integrates with SOSA compliance checker for security review
• Secured: Prompt injection scanning on all extracted text samples
• Agents: Supports multi-turn analysis with refinement iterations

Audit Report Structure

{
  "audit_id": "aud-2026-0329-001",
  "created_at": "2026-03-29T12:00:00Z",
  "scope": {},
  "baseline_metrics": {},
  "pattern_analysis": {},
  "recommendations": [],
  "compliance_status": "APPROVED",
  "next_review": "2026-06-29"
}

Success Criteria

• All recommendations achieve at least 10% token reduction
• Compliance checks pass 100% (no security/audit failures)
• Implementation priority matrix clearly ranked
• Report includes both quick wins and strategic optimizations
• Baseline metrics captured for post-implementation measurement