debrief

把面试或练习的口述复盘记录并入知识库，附 AI 置信评估。

把语音面试或练习的口述复盘（比如 `=== DEBRIEF ===` 块）收进知识库，生成会带 ai_confidence 的会话记录和凭证。用在你把面试回放或评分结果贴上来要入库、归档或追踪进步时。流程会做首次接触检查并保存评估与来源，方便后续检索和持续改进。

# /debrief — record verbal assessment results

Input: the interviewer's `=== DEBRIEF ===` block (preferred) or the user's own account.

**First-touch (run this first):** `.venv/bin/python -m brain first-touch debrief`. If it
prints a paragraph, this is the user's first debrief — open your reply with that text
verbatim, then continue. If it prints nothing, skip it and proceed.

1. Sanity-check before trusting it: levels must be plausible against the rubric and
   the justifications must contain actual quoted answers. If a claimed level has no
   supporting quote, downgrade to what the quote supports and say so. Map topic names
   onto the existing vocabulary (check current topics; don't mint