Description
Coding agents keep code running through round after round of changing requirements, but it grows bloated and hard to maintain. SlopCodeBench is a benchmark measuring code erosion under iterative specification refinement.
It simulates requirements evolving as in real development and quantifies how agent-written code degrades.
Iterative:Many rounds of changes.
Quality:Erosion quantified.
Comparison:Across agents.
It simulates requirements evolving as in real development and quantifies how agent-written code degrades.
Features
Iterative:Many rounds of changes.
Quality:Erosion quantified.
Comparison:Across agents.
