Do Context Files Help Coding Agents? A Two-Agent Ablation Study on Real Repositories
arXiv:2607.27250
The paper investigates whether providing persistent context files (like AGENTS.md) improves the performance of AI coding agents, finding no measurable benefit across Claude Code and Codex on real repository tasks.
Abstract
Persistent context files (AGENTS.md, CLAUDE.md) are standard practice for guiding AI coding agents, yet evidence for their effectiveness is contradictory. We present a controlled ablation of context-injection strategy across two frontier agents (Claude Code and Codex), 17 real tasks from 3 repositories (15 shared + 2 Codex-only), and 288 evaluated runs with gold-test evaluation. Context strategy does not measurably move correctness on either agent (bounded to <=10-15pp via equivalence testing). A failure-mode triage reveals why: agents fail on implementation skill---feature design, pattern selection, exact wiring---not missing repository knowledge that a context file could supply; a manipulation probe confirms the real AGENTS.md never converts a near-miss to a pass on either agent. We further show that borderline task difficulty is agent-specific (Spearman rho=0.75), offering a candidate explanation for prior contradictions: single-agent studies draw tasks from different agents' informative bands. We release all code, data, and analysis.