token-optimizer

To compress prompt context to fit a strict token budget, this tool applies pre-delegation compression to 500 tokens or fewer and post-delegation compression to 150 tokens or fewer, ensuring compact, cost-efficient reasoning before Engram storage.

To select the right Claude model tier and optimize prompt cost, this tool assesses task type, context size, and output requirements, selecting Haiku by default and escalating to Sonnet or Opus only when needed.

The best way to prevent token budget overruns in large delegation systems is to invoke a token optimizer at the start of every delegation to compress context and determine the initial model tier before downstream processing.

Yes, you can use token compression to reduce output size before Engram storage by applying post-delegation compression that condenses outputs to 150 tokens or fewer, ensuring deterministic routing and compact storage.

You should escalate from Haiku to Sonnet or Opus when the token optimizer assesses that the task type, context size, or output requirements demand a more advanced model tier for accurate results.

Yes, the token-optimizer works without external dependencies for cost optimization, applying deterministic routing decisions and context compression entirely within your delegation workflow.