Contributing / 参与贡献

中文

感谢你测试 DSH Passive Memory。当前最需要的是可复核的行为反馈，而不是完整私人 对话或主观分数。

提交 Issue 前请：

1. 确认版本、Harness commit、Node.js、模型和 recall mode；
2. 用合成文本或脱敏摘录描述“期望想起什么”和“实际想起什么”；
3. 说明问题发生在自动召回、L1 线索、邻域展开还是原始搜索；
4. 移除姓名、联系方式、密钥、完整 Prompt、数据库与未公开文件路径；
5. 可以提供内容无关的计数、工具调用顺序、耗时和错误类型。

代码改动应保持 L0 权威、L1 可重建、scope 严格隔离，并为新行为增加测试。 不要让召回消息或记忆工具输出重新进入 L0。

English

Thank you for testing DSH Passive Memory. Reproducible behavior reports are more useful than private transcripts or a single subjective score.

Before opening an issue:

1. record the plugin version, Harness commit, Node.js version, model, and recall mode;
2. describe expected and actual memory behavior with synthetic or redacted text;
3. identify whether the failure occurred in passive recall, an L1 clue, neighborhood expansion, or raw search;
4. remove names, credentials, complete private prompts, databases, and private paths;
5. prefer content-free counts, tool order, timings, and error categories.

Code changes must preserve authoritative L0 evidence, rebuildable L1 views, and exact scope isolation. New behavior needs tests, and recall/tool projections must never be re-ingested into L0.