Microsoft and UC Berkeley release TokenEconomizer to cut coding agent costs by up to 30.9%

TokenEconomizer achieves savings through three mechanisms: micro-skills (extracting reusable facts from traces), tool pruning (masking unused tools after three consecutive omissions), and failure-loop intervention (detecting and redirecting stuck agents). The plugin is Apache 2.0 licensed and runs locally, storing learned data in ~/.tokeneconomizer. While it improves Pass@1 accuracy in four of five settings, it reduces it by 1.0 point on Opus 4.8 in one case. The researchers note savings build over time, with all repositories becoming cheaper by the sixth task.