MimicHunterZ/dsh-agent-compact

Plugin插件 Native原生 ⭐ 4 NOASSERTION Sessions & Context会话与上下文

DSH插件,用于代理驱动的跨度压缩:将选定的对话跨度压缩成自写的检查点,而非官方的以头部为锚点的全上下文扫描。

Project Overview项目介绍

This is a native context compression plugin built exclusively for DeepSeek Harness (DSH). Unlike DSH’s default full-context compression that starts compressing from the beginning of the conversation, this plugin lets the DSH agent autonomously select a specific span of conversation to compress. To install the plugin from the official registry, run dsh plugin --profile <profile-name> add @mimichunterz/agent-compact, then restart your DSH profile to activate the tool. It preserves the important opening task requirements and the most recent context, minimizing information loss that occurs with default full-context compression. It also supports installing from a local checkout by running the same command with the path to your cloned repository instead of the registry package name.

The plugin is designed for DSH users who want more granular control over context compression. There are three common scenarios where the plugin is particularly useful: after completing a task step, after finishing a debugging session that went down the wrong path, or when the original opening requirements have gone stale. The agent selects the target span by specifying start and end anchors, and the plugin supports unique prefix matching that works with both full-width and half-width CJK punctuation. Before replacing the selected span with the agent-written checkpoint, the plugin archives the full raw text to a local spill file so it can be retrieved later if needed.

After installation, you can configure two main parameters through your profile’s cordis.patch.yml file. The autoArchive toggle enables or disables automatic archiving of raw compressed spans, which defaults to true, and volumeNudgeTokens sets the token threshold for nudging the model to use compression, which defaults to 50000. To uninstall the plugin, run dsh plugin --profile <profile-name> remove @mimichunterz/agent-compact and restart your DSH profile. If you added a custom mount entry in dev mode, you need to remove that entry as well. The plugin is built and verified for DSH version 0.1.1-rc.2, it only allows one compaction per session at a time, and it is released under the open source MIT license.

这是一款专为 DeepSeek Harness 开发的原生会话上下文压缩插件。不同于官方默认从会话开头进行全上下文压缩的机制,它允许代理自主选择需要压缩的特定会话区间,将已完成不需要保留的中间内容替换为代理自己编写的检查点,保留重要的开头任务要求和最新的上下文内容,减少压缩带来的信息损失。

插件面向需要更精细控制会话压缩的 DeepSeek Harness 用户使用。典型适用场景包括:完成某个任务步骤后压缩已完成内容、结束错误的调试或研究方向后归档相关讨论、开头需求过时后压缩过时开头并重述当前意图。代理通过指定起止锚点选中区间,支持唯一前缀匹配,兼容中英全半角标点,压缩前会先归档原始内容方便后续检索。

你可以通过 DSH 官方命令行安装:执行 dsh plugin --profile <你的配置文件> add @mimichunterz/agent-compact 后重启配置即可使用。卸载同理,执行对应 remove 命令后重启。插件配置支持调整自动归档开关和压缩提醒阈值,兼容 DSH 0.1.1-rc.2 及以上版本,采用 MIT 许可证开源。

Pre-install check安装前体检Compatibility · Security兼容性 · 安全性 1 warning1 项注意
  • Only 4 stars - very few users, little community feedback星标只有 4,几乎没人在用,遇到问题缺少社区反馈
DSH walks through these 9 checksDSH 会逐条核对这 9 项

Compatibility兼容性

  • DSH, Node, OS and profile requirementsDSH 版本 / Node 版本 / 操作系统 / profile 是否满足要求
  • External dependencies and runtimes (Electron / Python / Docker, ...)外部依赖与运行时(Electron / Python / Docker 等)是否齐备
  • Conflicts with installed plugins: command names, skill / tool names, ports, duplicate MCP registration与已装插件是否冲突:命令名、skill / tool 重名、端口占用、重复 MCP 注册

Security安全性

  • Repo matches the facts registered here; archived or abandoned?仓库是否与页面登记一致,是否归档或长期停更
  • Safety of preinstall / install / postinstall and install.sh / setup.ps1preinstall / install / postinstall 与 install.sh、setup.ps1 是否安全
  • curl|bash, download-then-execute, obfuscation, unrelated domains → stop immediatelycurl|bash、下载即执行、混淆代码、无关域名 → 立刻停止
  • Typosquatting or unmaintained packages among the new dependencies新增依赖里有没有 typosquatting 或无人维护的包
  • Requested permissions vs. what the feature actually needs申请了哪些权限、是否超出功能所需(filesystem / network / shell / clipboard)
  • Any sudo / admin requirement, plus uninstall and rollback是否要求 sudo / 管理员权限,以及卸载与回滚方式

Anything uncertain must be marked unknown with a note on how to confirm it. This site's signal screen is a static snapshot, not a security audit.拿不准的必须标「未知」并说明要我怎么确认。本站的信号筛查是静态快照,不能替代安全审计。

Or use CLI install (for developers)或使用命令行安装(适合开发者)

CLI Install命令行安装

dsh plugin --profile web add @mimichunterz/agent-compact

把 MimicHunterZ/dsh-agent-compact 加入你的 DSH 配置(web profile)即可启用。

READMEREADME

agent-compact

English · 简体中文

Context compression for DeepSeek Harness: lets the agent autonomously call context_compact to compress a span of the conversation it chooses — the finished, no-longer-needed middle — and replace it with a checkpoint the agent writes itself.

Why

Compaction is normally a full-context sweep: the official engine only ever compresses from the start of the conversation, so the opening's task plan and direction are partially lost along with the compressed information. context_compact compresses only the span the agent selects — a finished step, a debugged log exchange, an off-track discussion — while the important opening and the recent context stay intact. Span compaction keeps information loss from compaction as small as possible — like human memory, the middle is not compressed indiscriminately: what can be summarized is consolidated into a checkpoint, and important details stay word-for-word — the agent decides what is truly dead, and only that gets condensed.

Typical moments to use it:

  • a task step is done — compress it, keep the remaining steps and the active instruction live;
  • a bug hunt or a wrong research direction is over — compress that exchange into a short "what went wrong / root cause / fix" note;
  • the opening requirements are stale — compress the start and restate the current intent.

What it does

  • The agent picks the span via startAnchor / endAnchor (unique-prefix matching, CJK punctuation-width tolerant) and passes a required summary — the Markdown checkpoint it wrote itself.
  • The raw span is archived to the spill store first (~/.dsh/spill/session-<hash>/<hex>-<seq>.txt, sequential naming, restart-safe); the path is echoed in the shadow message so the model can read the raw text back.
  • The host engine runs the stock transaction — boundary validation, tool-pair balance, surface replacement — with no separate LLM summarizer request.

The tool call itself happens inside the agent's normal turn and is billed like any other turn; what is avoided is only the extra summarizer request the official engine would make for the same span.

Showing the opening section of the README — the full document lives in the repository以上为 README 开头摘要,完整文档在仓库内 · View the full README on GitHub →在 GitHub 查看完整 README →

← 上一个 Prev dsh-memory-toolkit 下一个 Next dsh-token-usage →