We describe a system for the CLPsych 2026 shared task on post-level identification of adaptive and maladaptive self-states. The system addresses subelement classification (Task 1.1) and presence rating (Task 1.2) with a retrieval-augmented in-context learning ensemble of two open-weight LLMs (Qwen3.5-27B and Mistral-Small-3.2-24B-Instruct) and a three-call prompt decomposition (unified, adaptive-focused, and Affect-focused extraction). Outputs are merged across models via deterministic aggregation with element-selection strategies tuned per subtask. The system placed 2nd of 17 on Task 1.1 (subelement Macro F1 = 0.441) and 5th of 17 on Task 1.2 (Avg RMSE = 0.994).