This paper presents ComplianceAgent, the legal‑compliance sub‑agent of the COMPASS multi‑agent framework. The agent relies on a large language model (LLM) to evaluate texts along four dimensions: digital sovereignty, carbon impact, compliance, and ethics. The model scores a text against a fixed rubric and grounds its judgment in a reference legal corpus. When the score falls at or below a critical threshold, a corrective chain is triggered, consisting of self‑healing rewrites, a compliance label, an explainability report, and escalation to a human reviewer. We describe the model’s role at each step, the hierarchical organization of the agent’s memory, and how the same implementation can be pointed to different reference corpora. Evaluation is conducted in a Canadian and Quebec context on a corpus of thirteen legal instruments covering federal, provincial, international treaties, and municipal documents of the City of Trois‑Rivières. A hand‑annotated gold standard of fifteen texts is used, combining classification metrics, repeated‑query stability tests, retrieval precision, and a composite score. Results show that the agent perfectly separates conforming from non‑conforming texts, its scores remain stable under repeated queries, and the index retrieves relevant passages. Review