Skip to content

docs(lark-drive): add answer-oriented search refinement #773

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Draft
BytedanceSearch wants to merge 1 commit into
base: main
Choose a base branch
from
+6 −0
Draft

docs(lark-drive): add answer-oriented search refinement #773

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 6 additions & 0 deletions skills/lark-drive/references/lark-drive-search.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -163,6 +163,12 @@ stdout 的 JSON 输出不受影响。`open_time` / `create_time` 不做 snap。
## 决策规则

- **和 `docs +search` 的选择**：优先使用 `drive +search`（本指令），不要再用 `docs +search`。`docs +search` 进入维护期、后续会下线。
- **答案型检索循环**：当目标是回答问题而不是简单列结果时，不要只跑一次原始 query 后直接收敛。至少按以下顺序做一轮 blind refinement：
1. 保留业务实体、时间、数值、产品名等硬约束，去掉“怎么/是否/为什么/近期情况”等问句噪声后重搜。
2. 对问题类型补充泛化检索词：案例类加“客户案例/最佳实践”，政策类加“政策/售卖/规则”，数值类加“指标/目标/准确率/金额”，对接类加“集成/同步/打通”。
3. 如果 top 结果标题强相关但未被读取，继续 fetch 强相关候选；如果 top 结果明显是评测集、Case 分析、Prompt/Debug 记录、标签标注样例、答案汇总等过程材料，只能在回答或 trajectory 里显式标记其性质，不能自动跳过、隐藏或降权。
4. 评测/指标类 query 不要一概排除“评测”文档：`Golden Set`、准确率评测方案、指标结果页可能就是答案来源；即使是会泄露评测过程或样例答案的材料，也必须保留可见轨迹，由 Judge 或用户判断是否污染。
5. 最终答案必须回填关键槽位：数值题给数字和单位，时间题给日期，是否题给 yes/no 结论，政策题给适用范围和限制条件。缺槽位时继续重搜或明确说明未找到。
- **身份快捷方式**：只要用户说"我创建的"，直接 `--mine` 即可，不需要先查 contact 拿 open_id。
- **时间维度选择**：
- "我编辑的"、"我修改的" → `--edited-since` / `--edited-until`
Expand Down
Loading