Background memory consolidation (auto dream) for DeepSeek Harness: periodically reviews recent sessions and existing memories to dedupe, refresh stale entries, remove contradictions and rebuild the MEMORY.md index, with hybrid/volume auto triggers, /dream command and a settings-panel config card.
Install
# from a prebuilt release tarball
dsh plugin --profile web add "https://github.com/yihefeikong-rgb/dsh-cc-haha-dream/releases/download/v0.1.0/dsh-dream-0.1.0.tgz"
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:yihefeikong-rgb/dsh-cc-haha-dream
Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).
README
This plugin publishes its README in Chinese only.
DSH 自动做梦插件:后台定期回顾近期会话与已有记忆,整合去重、更新过时、删除矛盾、修剪重建索引。
复刻 CC-HAHA 的 autoDream(后台记忆整合)行为,
与已安装的 dsh-memory 并存——dsh-memory 负责"随手记"(extractMemories),dsh-dream 负责"睡前整理"。
[!IMPORTANT] 本项目不是 CC-HAHA 官方组件,也不隶属于 CC-HAHA 或 DeepSeek。自动做梦的 门控链与整合流程参考 CC-HAHA(见 THIRD_PARTY_NOTICES.md), 插件代码使用 DSH 原生能力(会话查询、子代理、命令、工具、设置)独立实现。
[!IMPORTANT] 独立插件,不修改 dsh-memory;与它共享同一个
~/.dsh/memory/目录,文件格式与MEMORY.md索引完全兼容。
功能
- 自动定时:默认 24 小时 + 积累 5 个会话后,后台静默执行一次记忆整合(五重门控,不阻塞对话)。
从未做过梦(没有
.dream-lock)时不会冷启动即触发:启动后的首次检查只把锁盖成"当前时刻", 从此刻开始计minHours,之后的判断仍走原有门控。 - 手动触发:
/dream命令,或 agent 调用memory_dream工具,立即整合一次(任何时候都可手动触发,不受上面的冷启动守卫影响)。 - 整合内容(对齐 CC-HAHA autoDream 4 阶段,执行用 fork 子代理):
- 插件 fork 一个子代理(模型路由见下;可用
provider/model/maxTokens覆盖) - 子代理按 4 阶段提示词(定向 → 收集近期信号 → 整合 → 修剪索引)输出动作 JSON
- 插件在 host 侧安全落盘:合并重复(
merge/update)、修正过时/删除矛盾(delete)、 从近期会话提炼新记忆(create)、重建MEMORY.md索引(按## <项目>→### <类型>分节, 与 dsh-memory 的输出逐字节一致)
- 插件 fork 一个子代理(模型路由见下;可用
- 安全:所有写操作限制在记忆目录内;动作整批预检通过后才开始写入;删除/覆盖前自动存
.history/历史版本, 索引在原子覆盖前也会快照到.history/_index/(保留最近 20 份);LLM 输出严格 JSON 校验;.dream-lock防并发。
安装
dsh plugin --profile web add C:\Users\huangxinyang\.dsh\plugins\dsh-dream
# 或从源码目录:
dsh plugin --profile web add D:\claude\...\dsh-dream
安装后重启 DSH:
dsh web
配置
有两种方式,效果一致(设置面板写入的是用户层,覆盖 patch 默认值;applies: restart,重启后生效):
- 设置面板(推荐):打开 DSH 设置 → 插件 → 插件配置 选项卡 → 自动做梦 卡片,直接编辑并保存。
卡片字段含说明与"已覆盖/恢复默认"标记;保存后写入
~/.dsh/settings.yaml的dsh-dream:段。 - patch 文件:编辑
cordis.patch.yml的默认配置。
| 选项 | 默认 | 说明 |
|---|---|---|
enabled |
true | 后台自动做梦总开关 |
minHours |
24 | 距上次整合最小间隔(小时) |
minSessions |
5 | 距上次整合后最少新会话数 |
triggerMode |
hybrid | hybrid=时间或内容量先到先触发 / gates=仅固定间隔 / volume=仅按内容量 |
minVolumeChars |
30000 | 内容量触发阈值(近期待整合会话文本估算字符数) |
scanIntervalMinutes |
10 | 后台门控检查间隔(分钟) |
scopes |
auto | auto=近期活跃项目+通用 / all=全部 / current=当前项目+通用 |
maxDigestChars |
60000 | 近期会话摘要上限(字符) |
maxMemoryChars |
80000 | 记忆正文参考上限(字符) |
maxTokens |
8192 | 子代理整合输出上限;0=不传,由宿主/模型决定 |
agentTimeoutMs |
900000 | 做梦子代理总超时(毫秒) |
agentTools |
[](= read/grep/glob) |
做梦子代理可用的只读工具;宿主不支持时自动退回零工具 |
provider / model |
huoshang / deepseek-v4-flash-ga-260731 |
由 cordis.patch.yml 固定走火山(用户偏好,勿继承官方 deepseek-official);在设置面板清空则回退为"继承会话模型" |
manualOnly |
false | true=只允许手动/工具触发 |
storageDir |
~/.dsh/memory |
记忆存储目录(与 dsh-memory 共享) |
设置卡片属于插件浏览器端(client bundle)。新增或更新设置卡片后需重启 DSH, 让宿主重新扫描插件的
dsh.client声明(pkgMeta缓存只在进程启动时刷新)。
使用
- 手动整合一次:发送
/dream(可选all/current指定作用域范围)。 - agent 主动触发:调用
memory_dream工具。
整合结果会作为一条系统消息出现在当前会话;日志以 [dsh-dream] 前缀输出。
与 CC-HAHA 的关系
参考 CC-HAHA src/services/autoDream/ 的门控链与整合流程(autoDream.ts / consolidationPrompt.ts /
consolidationLock.ts),使用 DSH 原生能力(ctx.sessionQuery 读会话、ctx.subagents 派生子代理
(spawn → 不可用时回退 fork)做整合推理、ctx.commands / ctx.tools 入口)独立实现;
ctx.llm 仅用于枚举已注册 provider 以校验模型路由。
License
MIT
Links
More in this category
vectorize-io/hindsight#coding-agents★ 49309
Hindsight, agent memory that learns: long-term project memory with auto recall and retain, knowledge pages, deep reflection, and per-repo memory banks.
volcengine/OpenViking#examples/dsh-memory-plugin★ 39849
OpenViking memory and context bundle for DeepSeek Harness: pre-step auto-recall and profile injection, session capture, `viking://` URI guarding, and recall/write memory tools backed by an OpenViking server.
agentscope-ai/ReMe#dsh★ 3591
Connects DeepSeek Harness to ReMe's local-first, self-evolving personal knowledge base: automatically captures completed main-agent conversations as user-owned Markdown memory, searches conversations and source material through reme_search with BM25, optional embeddings, and wikilink expansion, and schedules daily memory consolidation.
zilliztech/memsearch#MemSearch★ 2743
Shared Markdown memory for DSH and other coding agents, with automatic capture, pre-step context injection, searchable recall, and memory-to-skill self-evolution through a review panel.
vshulcz/deja-vu#extensions/dsh★ 1191
Reads the session files thirty-three other coding agents on this machine already wrote — Claude Code, Codex, Cursor, VS Code Copilot Chat, opencode, OpenClaw, Hermes, Kimi, Cline, Zed and more — including sessions from before it was installed: six tools (deja_recall, deja_session, deja_blame, deja_fix, deja_how, deja_remember), a /deja command, and optional automatic recall added to the runtime context. Local BM25 index, no LLM, no embeddings, no network (dsh plugin --profile web add dsh-deja).
adoresever/graph-memory★ 647
Traceable, searchable cross-session memory for DeepSeek Harness — conversation knowledge as typed graph nodes (TASK/SKILL/EVENT) and typed edges.
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.