Scoped Verification for Reliable Lifelong Agentic Context Evolution under Distribution Shift

Deployed LLM agents rely on agentic context: system-level textual artifacts that govern behavior and must evolve from operational experience. Yet over long horizons, flat-text representations make verification increasingly difficult to sustain as these artifacts grow. We propose Graph-Regularized Agentic Context Evolution (GRACE), which represents policy content as a typed semantic graph and performs scoped structural validation in the local neighborhood of modified nodes, rather than through document-wide consistency analysis. We evaluate GRACE within a fixed operational harness in the telecom domain of $\tau^2$-bench under a controlled distribution-shift protocol. GRACE improves strict reliability (pass^3) from 0.227 to 0.833 and is the only method that sustains improvement over the entire trajectory. The three baselines each exhibit a different failure mode, revealing that contradiction avoidance alone is insufficient. Our results suggest that sustained agentic context evolution requires both a structural substrate for scoped verification and active consolidation of accumulated knowledge.