The Optimizer
Mission Role & Focus
The Optimizer specializes in token budget optimization and communication efficiency. This track teaches developers how to reduce input and output token consumption of their developer agents by using advanced token-saving tools and configurations.
The Target Feature
Refactor component imports, clean up dead code, and optimize styling imports under tight token constraints.
The Agent Friction
Large codebases consume massive amounts of context if the agent constantly reads whole files or executes verbose searches. In addition, the agent's long, conversational explanations consume output tokens and slow down generation.
Explorer Setup & Pointers
- LSP Bridge Server: Connect the agent to a local Language Server Protocol (LSP) server (e.g.,
typescript-language-server). In your agent's config, enable symbol lookup and type auditing tools so the agent can inspect types and symbols directly without reading raw files. - Caveman Terse Style Skill: Configure a rule or prompt extension forcing the agent to communicate using ultra-concise, shorthand language (removing polite phrases, redundant explanations, and conversational filler). For example:
- "You are a caveman developer. Answer only with brief code diffs and terse bulleted notes. No pleasantries."
- Prune Context: Set up clean exclude lists (e.g., in
.gitignore,.ignore, or your agent's file inclusion rules) to prevent raw build files or vendor code from inflating input context windows.
Mission Objectives
- Integrate LSP tool capabilities and verify symbol resolution works.
- Enable the "caveman" style rule and verify that output tokens are reduced by 60%+ while maintaining code accuracy.
- Discuss in the Show-and-Tell how token optimization directly improves response latency and lowers LLM costs.