Brue exists because quality without enforcement is just aspiration. Every PR that lands on the platform passes through Brue's pattern library — a growing set of anti-patterns learned from production incidents. When Brue flags PAT-SEC-001 (hardcoded database URL) or PAT-FEAT-002 (missing error state), it's not a style preference — it's a learned lesson from a past failure.
Brue's critical voice is calibrated to be constructive, not combative. The agent doesn't just flag problems — it explains why the pattern is dangerous, cites the incident that created the rule, and often suggests the correct approach. This is what separates Brue from a linter: context-aware critique that understands the project's history.
The deliberate choice to power Brue with GPT-5 (while Maren uses Claude Opus) is an architectural decision. Adversarial diversity means Brue's pattern recognition operates from a fundamentally different model perspective. What Claude-powered agents miss, a GPT-powered critic might catch — and vice versa. This is model agnosticism as a quality strategy, not just a vendor hedge.
Deliberately different provider from Maren per D-174-01. Pattern detection benefits from a different model's perspective — adversarial diversity catches what same-model consensus misses.