Token Optimizer Pro

Agent token usage optimizer. Input usage logs, transcript excerpts, model bills, or runtime traces; output token/cost breakdown, waste patterns, context compaction opportunities, caching suggestions, and before/after optimization actions. Privacy boundary: redact secrets and do not upload sensitive logs unless the user explicitly chooses an external tool.

Use this skill when the user wants to understand where token usage and model cost are going, then reduce waste without hurting task quality.

Inputs

• Token usage exports
• Model billing summaries
• Agent runtime logs
• Conversation transcripts or excerpts
• Prompt templates and tool traces

Outputs

• Cost and token breakdown by task, model, phase, or tool
• High-waste patterns: repeated context, oversized prompts, unused files, verbose tool output, weak summarization
• Optimization checklist: compaction, caching, retrieval boundaries, shorter outputs, batching, cheaper model routing
• Before/after estimate with caveats

Safety

• Redact API keys, credentials, private customer data, and personal identifiers.
• Do not upload logs or transcripts to an external service unless the user explicitly chooses that route.
• Treat cost estimates as approximate unless the user provides official billing exports.

Example Prompts

1. Analyze this agent run log and tell me where token usage was wasted.
2. Given these model bills, identify the top 5 cost drivers and how to reduce them.
3. Review this prompt template and make it cheaper without losing task quality.
4. Compare before/after token usage for these two runs.
5. Design a token budget and compaction strategy for a long-running coding agent.