Large Language Models (LLMs) can translate natural‑language optimization tasks into solver‑checked answers, yet current agents are brittle under narrative rewrites and often store reusable experience only at the instance level. We introduce \textsc{ArchOR}, an archetype‑centric skill agent that builds on \textsc{OptSkills} by adding two retrieval‑focused modules. \textbf{Centroid‑Aware Representation Learning} (CARL) refines the ordering of cached skill representations via a low‑rank prototype residual. \textbf{Signature‑Guided Retrieval} (S\textsuperscript{3}R) attaches open‑dimensional structural signatures to skills, prioritizes candidates by weighted Jaccard overlap, and bounds semantic re‑ranking to a top $K$ set. A baseline‑preserving arbitration rule decides when the structural route should replace the original selector. Using the same DeepSeek‑V4.1‑Flash backbone and released skill library, \textsc{ArchOR} achieves $65.00\%$ Micro‑Avg. Pass@1 across five benchmarks, improving over \textsc{OptSkills} by $1.64$ pp Micro‑Avg. and $2.68$ pp Macro‑Avg.\n\nReview: ArchOR’s combination of prototype‑aware representation and structural‑signature retrieval markedly boosts the robustness and overall performance of skill reuse, highlighting the promise of structure‑aware mechanisms in LLM‑driven optimization modeling.