Learning Meta-Skills for Agent Harness Design in Test-Time AI4AI
Frozen Builder models learn reusable Meta-Skills that improve agent harnesses by 8.95 points on Harness-Bench and NewtonBench.
3 stories, newest first · one card per story
Frozen Builder models learn reusable Meta-Skills that improve agent harnesses by 8.95 points on Harness-Bench and NewtonBench.
Empirical study of coding agent harnesses across 176 settings shows context management prevents overflow failures and bash-only interfaces cut costs for capable models.
Empirical study of 176 matched coding-agent harness configurations shows context management cuts overflow failures while planning and action-space choices trade accuracy against cost.
Today 1.6K readers · 2.5K views · 779 reads · all time 189K views