DeepSeek Harness Plugin Hub

Publish and manage complete Harness Profiles. Discover Plugins for your next setup.

Explore

PluginsPresetsDocsNews

Community

Publish a pluginContactReport an issue

Resources

Plugin Hub on GitHubDeepSeek HarnessSystem statusPrivacy notice
© 2026 DeepSeek Harness Plugin HubPowered byPaxTech

Independent and unofficial. Not affiliated with, authorized by, or endorsed by DeepSeek.

L123 Harness — DSH Plugin for DeepSeek Harness
DeepSeek Harness Plugin Hub
ProfilesPluginsCategoriesNewsDocsSign inManage Profiles
ProfilesPluginsCategoriesNewsDocsSign in
← Plugins
L

l123-harness

L123 Harness

L1-L2-L3 three-tier memory agent harness: gate approval, zero-judgement event log, weekly distillation, TDD execution loop. Claude Code plugin and DeepSeek Harness (dsh) bundle.

The plugin will be installed here. Keep web if you are unsure.

npx -y @deepseek-ai/dsh plugin --profile web add github:reatcat/l123-harness#9c52bfcb9ad4334ed38c4bcbd45dd8984e9d71cb
READMECompatibilityVersions

Compatibility and provenance

L123 Harness is published as l123-harness and currently resolves to version 1.1.0. The Hub verifies its manifest and preserves the exact installation source for reproducible installs.

DSH compatibility
*
Runtime surfaces
any
Release source
github
Registry updated
8/22/2026

Versions

1.1.0stable
8/22/2026

Related plugins

Loading related plugins…

Latest
1.1.0
DSH
*
HMR
Process restart
Tree shaking
Safe tree shaking not declared
Unpacked size
Unavailable
Files
Unavailable
Surface
any
License
MIT
Source
github
GitHub
★ 3
Weekly downloads
0
Last push
8/22/2026
View source ↗
README badge

Click the badge to copy Markdown for your README.

Do you maintain this Plugin?Claim benefit · Priority security scan

Verify the GitHub repository declared in package.json to manage this listing. After you claim it, Hub will prioritize a security scan of the current version and publish the result when it passes.

Claim this Plugin →
Report an issue

Related plugins

More verified plugins in memory-context.

Memsearch Dsh@zilliz/memsearch-dshMemSearch plugin for DeepSeek Harness: shared markdown memory across agents, with capture, pre-step context injection, memory-recall skill, and a skill-candidate review panel.Reme@agentscope-ai/remeReMe client and memory integrations for TypeScript agentsStratagate Dshstratagate-dshRecent conversations stay vivid. Older ones fade into summaries, not oblivion. StrataGate gives DeepSeek Harness six-layer, time-decaying memory, while lasting events and relationships settle into a knowledge graph. Bring your memories from other AIs withMeow Memorymeow-memoryCross-session project memory for DeepSeek Harness: seven-layer SQLite memory, first-turn snapshot injection, per-message keyword hits, memory_remember/search/project tools, automatic reflection with reflection-fold UI, and idle-triggered dream consolidati

README

l123-harness —— L1-L2-L3 三级记忆 Agent 底座

把一套经过实战检验的 agent 方法论打包成插件:Claude Code 插件 + DeepSeek Harness (dsh) 组合包,同一套内容双宿主分发。设计对齐「一切皆插件」: 宿主是内核,本插件只提供能力组件——上下文注入、门禁、事件记录、记忆提炼, 全部可单独理解、可整体装卸。项目知识不在插件里:插件装机后由项目自己积累。

English — l123-harness packages a battle-tested agent methodology as a plugin for both Claude Code and DeepSeek Harness (dsh): a three-tier memory system (L1 always-on CLAUDE.md / L2 index / L3 details) plus gate approval for protected files, a zero-judgement event log, weekly distillation, and a TDD execution loop. Install for dsh: dsh plugin --profile web add github:reatcat/l123-harness

四条原则(方法论内核)

  1. 状态放模型外:结论、计划、进度写盘,上下文只装当前任务需要的一小部分。
  2. 客观观察:测试红绿、命令输出、日志数据才是判据;推理不是证据。
  3. 内层不发散:spec 管计划、计划管步骤边界、TDD 管停止条件。
  4. 升级需审批:memory 随手写;升级为 knowledge / skill / L1 / L2 必须用户批准。

三级记忆

层载体特点
L1 常驻CLAUDE.md小而稳定的缓存前缀:原则、优先级、安全底线、路由表
L2 索引.agents/INDEX.md症状→一句规则→L3 指针;SessionStart hook 自动注入
L3 详情.agents/**knowledge / pitfalls / runbooks / current / events / inbox

组件清单

l123-harness/
├── .claude-plugin/plugin.json      # Claude Code 插件清单
├── package.json                    # dsh 组合包清单(dsh.bundle)
├── cordis.patch.yml                # dsh 组合层:插入本插件行
├── dsh/index.js                    # dsh 入口:注册技能 + 挂载官方 hooks 桥接
├── hooks/
│   ├── hooks.json                  # 事件接线(6 个生命周期事件)
│   ├── on-session-start.sh         # 记事件 + 注入 L2 索引
│   ├── gate-guard.sh               # PreToolUse 门禁:保护清单 + 一次性审批令牌
│   ├── on-post-tool.sh             # 工具报错 → tool_error 事件(热路径零解释器)
│   ├── on-pre-compact.sh           # 压缩前抢救 .agents/current/ 为 checkpoint
│   ├── on-stop.sh / on-session-end.sh  # 边界事件记录
│   └── lib-event-log.sh            # 事件库:不可变、按月滚动、绝不阻塞主流程
├── skills/
│   ├── init-harness/               # /l123-harness:init-harness 装机(含 init.sh)
│   ├── event-review/               # /l123-harness:event-review 周审提炼
│   ├── tdd/                        # 执行流的停止条件(红绿循环)
│   └── write-a-skill/              # 「连续成功 3 次 → skill 草稿」的落地工具
└── templates/                      # 装机材料:L1/L2/gate.conf 模板 + 5 个工作模板

安装

Claude Code

# 方式 A:本地开发/单机
claude --plugin-dir /path/to/l123-harness

# 方式 B:个人常驻(放进 skills 目录,自动加载为 l123-harness@skills-dir)
cp -r l123-harness ~/.claude/skills/

# 方式 C:git 仓库分发(本目录即单插件 marketplace)
# 他人执行:
#   /plugin marketplace add <git-repo>
#   /plugin install l123-harness

DeepSeek Harness(dsh)

本仓库同时是标准 dsh 组合包(package.json 声明 dsh.bundle + 根目录 cordis.patch.yml),纯 JS 免构建,git 安装无需 pnpm 构建授权:

dsh plugin --profile web add github:reatcat/l123-harness

挂载时自动完成两件事:

  1. 技能注册——4 个方法论技能注册到 ctx.skills,经标准 skill 工具 面向模型与用户可用(名称与 Claude Code 下一致);
  2. hooks 桥接——挂载官方 @deepseek-ai/dsh-hooks-claude-code, 原样运行 hooks/hooks.json(CC/DSH 双宿主同一份钩子脚本)。
hooks.json 事件dsh 桥接说明
SessionStart(L2 注入 + 事件)✅输出改为 JSON additionalContext,双宿主兼容
PreToolUse(gate-guard 门禁)✅matcher 已含 DSH 的 edit/write/str_replace_editor;exit 2 → deny
PostToolUse(tool_error 记录)✅纯观察事件
Stop(边界事件)✅纯观察事件
PreCompact(压缩抢救)⚠️ 桥接未实现配置被安全忽略,该能力在 DSH 暂不生效
SessionEnd(边界事件)⚠️ 桥接未实现同上

注:DSH 上 init-harness 生成的骨架目录约定(CLAUDE.md / .agents/) 与 Claude Code 完全一致,两个宿主可共用同一个项目的记忆骨架。

使用

/l123-harness:init-harness     # 新项目装机:生成 L1/L2/L3 骨架 + 门禁清单
(日常干活:L2 索引自动注入;门禁自动拦截受保护文件;事件自动记录)
/l123-harness:event-review     # 周审:读事件日志 → 候选进 inbox → 陪同分流

DSH 下同名技能经 skill 工具或用户命令调用(init-harness、event-review、 tdd、write-a-skill),日常自动化部分(L2 注入、门禁、事件记录)由 hooks 桥接在对应生命周期点自动运行。

纪律速查

  • 改受保护文件:先向用户说明 → 批准 → touch <项目根>/.claude/.gate-token → 重试。
  • .agents/inbox/ 干活时不得读取;候选不是结论。
  • .agents/events/ 只读不改,不可变。
  • 知识升级路径:context → memory(随手)→ inbox(周审)→ knowledge / L2(人批)。

许可

MIT