token-optimization

# neuron: token-optimization - triggers_when: seeking to save tokens or the context exceeds 60% of the limit - tier: T1 - budget: 300 tokens - eval: context <=60% and no repeated context -> threshold 1 - reward: pheromone +1 if eval >= threshold ## Procedure 1. Never repeat context: use memory/STATE.md, AGENTS.md and diagrams. 2. Prefer diagrams over text (5x fewer tokens). 3. Use short prompts with file references. 4. Always update memory/STATE.md when finishing. 5. Compact when exceeding 60% or changing phase. ## Checklist - [ ] Context in files, not repeated in the prompt - [ ] Diagrams for complex flows - [ ] One task per prompt - [ ] Explicit response format; memory/STATE.md updated ## Pitfalls - Pasting complete files when one section suffices - Re-explaining context already in memory/STATE.md - Not compacting on multi-day tasks