query

基于个人二脑做检索增强回答，区分用户已知信息与新增建议。

结合你的 second‑brain（个人笔记/知识库）做检索增强回答，把你已经知道的内容和我给出的新建议分开列明，适合问学习、项目或“我关于 X 已知什么”的场景。遇到索引过期或缺失时，会先尝试跑一次索引同步（brain ingest）再回答并说明已同步。特点是用 RAG 风格一边引用笔记一边给出可操作的新建议。

# /query — RAG answer over the knowledge base

**Fail soft (ux.md #8):** if search errors, returns nothing on a topic that
clearly has notes, or `brain status` would show pending changes, run
`.venv/bin/python -m brain ingest` yourself and retry once — then mention it
in one line ("synced the index first"). A deleted/stale index is a known
mode with a deterministic fix, never an error to show the user.

1. Search, more than once if the phrasing allows: run
   `.venv/bin/python -m brain search "<query>" -k 5` and, when useful, a second
   reformulation or `--domain`/`--goal` filtered pass.
2. Read the full top notes (not just snippets) — frontmatter matters: confidence,
   goals, dates tell y