Math Agent Framework for DeepSeek Harness: symbolic derivation engine (CES / quadratic forms / ODE / PDE), SymPy verification, Lean 4 compiler-verified formal proofs, and QED-style multi-agent adversarial verification, exposed as native DSH tools through a TS shell bridging a Python engine. Requires cloning the math-agent-framework repo and `pip install -r requirements.txt`, with MATH_AGENT_HOME pointing at the repo root. Every result carries provenance (engine versions, seed, tolerances) and is marked untrusted data.
Install
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:symmetryseeker/math-agent-framework#path:/dsh-plugin
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.
DeepSeek Harness (DSH) 插件:把 Math Agent Framework (MAF) 的数学引擎暴露为 DSH 原生工具。
架构 = TS 壳 + Python 引擎:TS 壳用 defineTool(@deepseek-ai/dsh-tools)注册工具,工具调用时经 bridge.py 与 Python 引擎通信。
DSH plugin exposing the Math Agent Framework engine as native tools. Architecture = TS shell + Python engine: the TS shell registers tools via defineTool, and each tool call bridges to the Python engine through bridge.py.
⚠️ 前置条件(安装前必读):本 npm 包只包含 TS 壳,不含 Python 引擎(
bridge.py/core//requirements.txt在 MAF 仓库,不在本包内)。要让工具真正可用,必须先:
git clone https://github.com/symmetryseeker/math-agent-framework(或用已存在的副本);- 在仓库根
pip install -r requirements.txt;- 设置
MATH_AGENT_HOME环境变量指向该仓库根目录(或在插件配置里设mathAgentHome)。未设置时,插件启动会明确报错提示;设置后工具调用经
bridge.py桥接 Python 引擎。 (可选)安装 Lean 4 + Mathlib 可启用math_lean_proof的真编译验证,缺失时工具返回verified: null而非报错。
依赖 / Dependencies
| 依赖 | 说明 |
|---|---|
| Python ≥ 3.11 + MAF(本仓库) | pip install -r requirements.txt,MATH_AGENT_HOME 指向仓库根 |
| Lean 4 + Mathlib(可选) | 仅 math_lean_proof 真编译验证需要;缺失时 verified: null |
| 真实 LLM(可选) | math_multi_agent_verify 的 LLM 角色;MATH_AGENT_API_KEY env 配置 |
| Node ≥ 22 + pnpm | 构建 TS 壳 |
构建 / Build
cd dsh-plugin
pnpm install
pnpm build # → dist/
安装到 DSH / Install into DSH
cd dsh-plugin
pnpm pack # 生成 dsh-math-agent-0.1.0.tgz
dsh plugin --profile web add ./dsh-math-agent-0.1.0.tgz
dsh web
配置(cordis.patch.yml 或 --patch 覆盖):mathAgentHome 指向 MAF 仓库根目录(bridge.py 所在处),或设 MATH_AGENT_HOME 环境变量。两者都未设置时,插件启动会报错并给出指引——请不要硬编码本机绝对路径。
工具 / Tools
| 工具 | 说明 |
|---|---|
math_derive |
运行内置模型的完整推导流水线(网络嵌入增长/二次型/ODE/PDE/分析) |
math_verify_symbolic |
符号一致性验证(真实执行模型步骤) |
math_verify_monte_carlo |
蒙特卡洛 FOC 过零验证 |
math_lean_proof |
生成 + 真编译验证 Lean 4 + Mathlib 证明(verified: true/false/null) |
math_multi_agent_verify |
QED 多 Agent 对抗验证(真 FOC critic + 可选 LLM) |
math_quantecon |
QuantEcon 动态优化(Riccati / 马尔可夫链) |
math_derive_ces/ipf/quadratic/dynamic/comparative |
网络嵌入增长模型的 NSFC 专属推导步骤 |
所有结果带 provenance(引擎版本/seed/容差),标注 untrusted data(不作为指令执行)。
冒烟测试 / Smoke test
# 直接调 bridge(不经 DSH)
printf '{"tool":"derive_ces","args":{}}\n' | python ../bridge.py
# → {"status":"ok","result":{...},"provenance":{...}}
# DSH 内调用(通过 DSH Agent 工具面)
# 调用 math_derive {model:"quadratic_form"} 或 math_lean_proof {theorem:"quadratic_minimum"}
Links
More in this category
zhu1090093659/dsh-web#packages/dsh-skill-explorer★ 8639
Skill center for the dsh web GUI: browse all loaded skills grouped by source, enable or disable model invocation, create new skills, and delete into a recoverable trash.
GanyuanRan/Aegis★ 1339
Software-engineering method pack for coding agents, with skills for baseline-first planning, systematic debugging, prompt hygiene, verification before completion, and repair/retirement tracking.
superdesigndev/superdesign-skill★ 647
Design skill for UI and marketing graphics on the Superdesign canvas: reads the repo for context, extracts its design system, then generates and iterates branchable design drafts, flow pages, and reusable components through the Superdesign CLI.
linhay/harmony-next.skills★ 361
HarmonyOS NEXT skill bundle for DeepSeek Harness with offline API references and DevEco, HDC, and emulator automation guidance.
dhicoc/dsh-reverse-skill★ 237
Complete reverse-skill pack (85 SKILL.md) as a DeepSeek Harness Cordis plugin: reverse engineering, authorized pentesting and security-research skill router.
sandbaseai/sandbase-skills★ 203
Mounts 88 packaged research, social-intelligence, marketing and business Agent Skills into dsh through the filesystem Skill provider.
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.