988hj7tczd-oss/dsh-math-olympiad

DSH skill bundle: competition math (IMO/Putnam/USAMO/AIME) solved with pure reasoning, adversarial verification in a fresh subagent, and calibrated confidence

Project Overview项目介绍

dsh-math-olympiad is a profile-style skill bundle built specifically for DeepSeek Harness, targeting competition mathematics problems drawn from IMO, Putnam, USAMO, AIME and similar contests. The package ships as a dsh.bundle declared in package.json, with a dsh.bundle.patch overlay that registers two cordis plugin lines: dsh-math-olympiad-locator, which exposes the bundled skills directory as a mathOlympiadSkills service located via import.meta.url, and dsh-math-olympiad-skill, an isolated providerName: math-olympiad filesystem provider with includeDefaultRoots: false to avoid double-scanning the base roots. Installation is performed with dsh plugin --profile <name> add ./dsh-math-olympiad against a profile that already includes @deepseek-ai/dsh-base, after which the bundle is launched under dsh --profile <name>.

The intended workflow runs a five-step loop documented in SKILL.md: first solve purely by reasoning using Pólya's six heuristics rewritten in Chinese, then strip the chain of thought so only the proof body remains, next dispatch a subagent with a fresh context to adversarially verify the proof against verifier_patterns.md, fall back to step one on a fatal gap for at most two iterations, and finally emit a tuple of solution plus verification verdict plus calibrated confidence (high / medium / or the honest "no confident solution" abstention with the sticking point). Optional LaTeX output is produced when check_latex.sh detects an engine, otherwise the flow degrades gracefully to Markdown math. The bundle is aimed at competition trainees, olympiad coaches, and researchers who want a rigorous process over a promised accuracy rate.

Distribution and limits are explicitly documented: the npm-released tarball is self-contained with the lib/ build already produced, so plain dsh plugin add ./dsh-math-olympiad-0.1.0.tgz works without any build approval; however, installing directly from git requires pnpm ≥10 and either adding allowBuilds: dsh-math-olympiad: true to the profile's pnpm-workspace.yaml or running the interactive pnpm approve-builds, because pnpm ≥10 refuses to run the prepare script of git dependencies by default. Verification is done with node tests/smoke.e2e.ts on Node ≥23.6 (or npx tsx on 22.6–23.5), exercising structure, trigger words, the five red-flag heuristics, the verifier protocol, the abstention path, and shell-script behavior with engine isolation. Limitations: correctness is not guaranteed since adversarial verification improves rigor but cannot eliminate errors, and subagent quality depends on the model plus disciplined context isolation. License is MIT, all prose is independently rewritten in Chinese, no proprietary upstream code is included, and dsh-safe-plugin-manager re-checks the bundle every three hours against the DSH STORE five-category review signals.

dsh-math-olympiad 是一个面向 DeepSeek Harness(DSH)的竞赛数学技能 profile bundle。它把 IMO、Putnam、USAMO、AIME 等题型封装为 SKILL.md 流程:纯推理求解 → 剥离思考链 → 新鲜上下文 subagent 对抗验证 → 校准信心输出 high/medium/no confident solution,并可选 LaTeX 编译 PDF。安装时通过 dsh plugin --profile <name> add ./dsh-math-olympiad 挂载,依赖 profile 中预先存在的 @deepseek-ai/dsh-base,由 package.json 的 dsh.bundle.patch 将 mathOlympiadSkills 服务注册到 skill-filesystem。

典型用法是用户以触发词发起一道竞赛题,主插件定位技能目录后启动五步循环:解题、剥链、subagent 对抗验证、致命缺口最多回退两轮、最终输出解 + 验证结论 + 信心等级,不确定时直接诚实地声明 "no confident solution" 并附卡点。适合希望以严谨流程而非承诺正确率处理高难度题目的研究者与竞赛训练者,以及需要对抗验证降低幻觉风险的用户。

依赖方面,发布形态自带 lib/ 构建产物可直接 add;git 直装需 pnpm ≥10 在 pnpm-workspace.yaml 中加 allowBuilds: dsh-math-olympiad: true 或运行 pnpm approve-builds,否则 prepare 脚本会被默认拒绝。LaTeX 编译非必需,缺引擎时 check_latex.sh 退出码 1 并提示安装,流程自动回退到 Markdown 数学。许可 MIT,DSH STORE 复检由 dsh-safe-plugin-manager 每三小时执行。

Pre-install check安装前体检Compatibility · Security兼容性 · 安全性 1 warning1 项注意
  • Only 2 stars - very few users, little community feedback星标只有 2,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 github:988hj7tczd-oss/dsh-math-olympiad

把 988hj7tczd-oss/dsh-math-olympiad 加入你的 DSH 配置(web profile)即可启用。

READMEREADME

dsh-math-olympiad

[!IMPORTANT] 依赖前置:相邻 dsh-src 检出(link: 依赖) 本项目在开发形态下使用 link: 依赖指向相邻的 DeepSeek Harness 源码检出(dsh-src), 与当前仓库保持同一父目录布局(<parent>/dsh-src)。克隆本仓库后:

  1. 先把官方 deepseek-ai/deepseek-harness 检出到与本仓库同级的 dsh-src/ 目录,并执行其 pnpm install && pnpm run build;
  2. 再按下方「安装」一节执行本仓库的 pnpm install --offline && pnpm build 与测试。 发布到 npm 的版本会尽量把 link: 依赖替换为 registry 真实版本;无法替换的内部包保持 link:,见各包 README 说明。

DSH 技能插件:竞赛数学解题(对抗验证 + 校准信心)。 面向 IMO / Putnam / USAMO / AIME 及各类数学竞赛题,流程为 "纯推理求解 → 剥离思考链 → 新鲜上下文 subagent 对抗验证 → 校准信心输出 (high / medium / 诚实 'no confident solution')",可选用 LaTeX 编译 PDF。

定位:这是严谨流程 + 诚实弃权的技能,不承诺正确率。 输出总是"解 + 验证结论 + 信心"三位一体;不确定就直说,不硬编答案。

目录结构

dsh-math-olympiad/
├── package.json                  # dsh.bundle 声明(发布为可安装 profile bundle)
├── cordis.yml                    # 开发期补丁覆盖层(scratch-plugin 风格,引用本机 src)
├── cordis.patch.yml              # 发布用 bundle 层(引用安装后的 lib 入口)
├── tsconfig.json                 # src/ → lib/ 构建配置
├── src/locator.ts                # Host 插件:提供 mathOlympiadSkills 服务(定位 skills/)
├── skills/dsh-math-olympiad/
│   ├── SKILL.md                  # frontmatter(触发词) + 5 红旗启发式 + 五步流程
│   ├── references/
│   │   ├── solver_heuristics.md      # Pólya 六板斧(中文重写)
│   │   ├── verifier_patterns.md      # 验证器失败模式清单(真实攻击面)
│   │   └── presentation_prompts.md   # 输出/讲解模板与数学排版约定
│   └── scripts/
│       ├── check_latex.sh            # TeX 引擎探测(无引擎给出各平台安装提示)
│       └── compile_pdf.sh            # LaTeX → PDF(两遍编译,退出码语义明确)
├── tests/smoke.e2e.ts            # 离线冒烟:结构/触发词/5 原则/subagent 协议/弃权路径/脚本行为
├── README.md
└── LICENSE                       # MIT

安装(发布形态:profile bundle)

技能以 @deepseek-ai/dsh-skill-filesystem 的 customSkillDirs 挂载 (PROMPT 指定的 DSH 侧集成方式):

cd dsh-math-olympiad 的父目录
dsh plugin --profile <name> add ./dsh-math-olympiad
dsh --profile <name>
  • 前置:profile 需含 @deepseek-ai/dsh-base(默认第一个 bundle,提供 skill 注册表与 @deepseek-ai/dsh-skill-filesystem 包)。
  • package.json 的 dsh.bundle.patch 指向 cordis.patch.yml;该层插入两个 行:dsh-math-olympiad-locator(把技能目录暴露为 mathOlympiadSkills 服务,构造时从 import.meta.url 定位包内 skills/)与 dsh-math-olympiad-skill(独立的 providerName: math-olympiad skill-filesystem 提供者,避免与 base 行冲突;配置 includeDefaultRoots: false,只扫本包的 customSkillDirs,不重复扫描 默认根)。
  • git 直装只取源码:prepare 负责构建 lib/(自包含 tsc,无需 monorepo 上下文)。但 pnpm ≥10 默认拒绝运行 git 依赖的 prepare, 首次 add 会失败——先允许构建再重跑 add:在 profile 的 pnpm-workspace.yaml 里加

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-cache-stabilizer 下一个 Next dsh-ssh-tui →