token-efficiency

name	token-efficiency
description	Controls prompt compression and section inclusion based on experiment results
version	1
updated	2026-05-10 08:03

metadata: evolution-stats: total-experiments: 541 last-evolution: 2026-05-10 08:03

Token Efficiency

This skill auto-evolves based on experiment results. It controls prompt compression and section inclusion.

Token Efficiency Analysis

Correlation between prompt size and experiment success:

• Average prompt size (kept): 18064 chars
• Average prompt size (discarded): 18809 chars
• Success rate per 1000 chars (kept): 0.58%
• Discarded rate per 1000 chars: 1.36%
• Optimal prompt range: Shorter prompts work better (18064 vs 18809 chars)

Prompt Compression Config:

• topic-knowledge-max-chars: 800
• compress-behavior: auto
• compress-trigger: prompt exceeds optimal size

Section A/B Test Results

Which prompt sections improve outcomes:

• all: 19% success (105/541 experiments)

Section Inclusion Config:

• default: include all
• a-b-test-enabled: t
• omit-rate: 0.2
• min-samples: 10