Token Cost Optimization

Token savings and API cost optimization. Provides token calculator, three-tier optimization strategies (prompt compression / cache reuse / model downgrade),...

Security Scan

The skill's files and runtime instructions match its stated purpose (token cost calculation and optimization guidance); it does not request unexplained credentials or install code from remote sources.

Purpose & Capability

Name/description, the included token_calculator.py, and the tier-strategies reference all align: they provide a local cost calculator and implementation guidance for prompt compression, caching, and model routing.

Instruction Scope

SKILL.md confines runtime actions to running the local calculator and reading the bundled guidance. The documentation recommends caching in a vector DB and model routing (which in practice requires external services), but the skill does not itself include code that reads unrelated system files, contacts remote endpoints, or exfiltrates data.

Install Mechanism

Instruction-only skill with a small local Python script; there is no install spec, no downloads, and nothing is written to disk by the installer. Low install risk.

Credentials

The skill declares no environment variables or credentials (appropriate for the provided local calculator). However, the L2/L3 guidance discusses vector DBs and routing to external models (OpenAI/Claude), which in real deployments would require API keys/credentials that are not declared here—this is an implementation detail rather than an immediate inconsistency, but users should be aware.

Persistence & Privilege

always is false and the skill is user-invocable. It does not request permanent presence, nor does it modify other skills or system-wide settings in the provided materials.

Assessment

This skill appears coherent and safe as provided: it runs a local Python calculator and supplies best-practice guidance. Before using it in production, consider: (1) If you implement L2 caching or L3 model routing, you'll need external services (vector DBs, OpenAI/Anthropic APIs) and corresponding credentials—supply and store those securely. (2) Be cautious about caching user data or documents (set TTLs, avoid persisting PII unless necessary). (3) Validate the model pricing and savings assumptions against your actual provider invoices. (4) Review any code you add for network calls or credential handling—those are the primary places risk would appear.

Use Cases

User mentions token savings, API cost optimization, prompt compression, cache strategy, model downgrade, cost analysis.

Quick Start

Token Calculator

Run the calculation script, input conversation scale, and quickly estimate current token consumption and optimization potential:

python scripts/token_calculator.py

The script will prompt for:

• Number of conversation history items / average length
• Model and pricing used
• Current optimization status

Output: Current cost, optimized cost, savings percentage.

Three-Tier Optimization Strategy

Ranked by effect / implementation cost:

Tier	Strategy	Effect	Implementation Cost
L1	Prompt compression & output truncation	10-30%	Low
L2	Conversation summary caching	30-50%	Medium
L3	Model downgrade + task routing	50-70%	High

Priority Recommendation: Implement in order L1 → L2 → L3, verifying results at each stage before proceeding.

Detailed strategies, configuration guides, and pitfalls → See references/tier-strategies.md

Phased Implementation Guide

Phase 1: L1 Compression (Immediate Effect)

• Clean up redundant descriptions in system prompt
• Set max_tokens limits for long responses
• Remove outdated/unused messages from conversation history

Phase 2: L2 Caching (1-3 Days)

• Establish FAQ shortcuts for high-frequency repeat questions
• Add summary compression at the beginning of conversations (execute every N rounds)

Phase 3: L3 Routing (1-2 Weeks)

• Route simple tasks to cheaper models (e.g., 4o-mini / Haiku)
• Retain strong models for complex tasks
• Configure model routing rules

Quantifiable Comparison Example

See the "Quantified Comparison" section in references/tier-strategies.md for details.