Configuration Smells in AGENTS.md Files Waste Tokens and Reduce Agent Efficiency

Researchers from the Federal Institute of Minas Gerais analyzed 532,000 files from 100 popular open-source projects, revealing widespread 'configuration smells' in AGENTS.md and CLAUDE.md files. The study identified issues such as Lint Leakage (62% of files), Context Bloat (42%), and Skill Leakage (35%). These problems lead to inefficient token usage and hinder agent performance. Developers are advised to limit configuration files to 200 lines to enhance model coherence. The findings suggest that many developers are unaware of these inefficiencies, which can lead to increased operational costs and reduced effectiveness of coding agents.

Key Points: • Configuration smells in AGENTS.md files waste tokens and degrade agent performance. • Lint Leakage, Context Bloat, and Skill Leakage were found in a majority of analyzed files. • Developers are encouraged to limit configuration file length to improve efficiency.