token-optimiser

What problem does it solve? Running every agent task on premium LLM APIs wastes tokens and money, while routing everything to weak local models produces poor results. This Skill scores each task for confidence and scope, then routes it to the cheapest viable tier — local GPU, free cloud models, or paid escalation — with governance and learning built in. ## Core Features & Use Cases - 3-Tier Budget Routing: Routes tasks to T1 local llama-server, T2 free cloud models (OpenRouter, NVIDIA NIM, 40+ free models), or T3 paid models based on confidence thresholds (T1 ≥80%, T2 ≥60%, T3 <60%). - BOA + ITIL Governance: A 6-seat Board of Advisors (Architect, Engineer, Researcher, Reviewer, Budget, Executive) with ITIL classification governs tier selection and automatic swarm delegation for non-trivial work (>3 files, >2 modules, >5 tool calls). - Learning Log Calibration: Every routing decision and outcome is appended to a JSONL learning log to calibrate confidence scoring over time. - Use Case: A developer asks the agent to fix a typo, write unit tests, and add OAuth2 authentication. The Skill routes the typo fix to a free T2 model, the tests to T2, and the security-sensitive OAuth work to T3 with a stripped prompt, logging each decision. ## Quick Start Ask the agent to route this task with the token optimiser, or invoke /token-optimiser before describing your task to get a Token Budget Plan and tiered routing decision.