Claude defaults to he/him for the user despite stated she/her, including in visible reasoning

What happens
Claude intermittently refers to me with he/him even though my pronouns (she/her) are stated — both in my "Instructions for Claude" profile and in the assistant's own persistent memory. It has happened in normal replies and, notably, in the model's visible reasoning/thinking, where it wrote "his vision" and "he'd promote it" about me mid-task.

Why it matters
This isn't a cosmetic slip. Being repeatedly read as "he" is alienating, and the neutral default (they/them) exists precisely to avoid it — so sliding to he/him is worse than either the correct pronoun or the neutral fallback. It recurs even after the correct pronoun is recorded in memory, which suggests the stored fact isn't reliably reaching generation (especially the reasoning channel).

Expected
Once pronouns are stated (profile or memory), Claude should use them consistently everywhere it writes — replies, visible reasoning, commit messages, generated docs. Where pronouns are unknown, it should hold the neutral they/them default and never infer gendered pronouns from a first name.

Environment
Claude Code CLI, model Claude Opus 4.8, Windows 11.