vibe-mathematics: A Multi-Agent Math Solving and Verification DSH plugin for DeepSeek Harness

chongcyrus/vibe-mathematics

Skills & AgentsVerified
Listed on 2026-08-14
Page last updated 2026-10-04

Vibe Mathematics is a multi-agent math solving and verification framework for DeepSeek Harness, offering four agent presets with resume, human intervention, and cross-validation.

vibe-mathematics is a multi-agent mathematical problem solving and formal verification plugin running in DeepSeek Harness, offering four agent presets that collaborate to automatically solve math problems and cross-verify conclusions. It addresses the issues of single-agent solvers getting stuck and conclusions lacking credibility, letting you drive the full loop of exploration, iteration, verification, and consolidation in natural language. Core capabilities include checkpoint resume, mid-way human intervention, progress reporting, and four generations of solving architectures, with v2 and v3 as the main recommendations and v4 as an experimental resident self-organizing research architecture.

How to Install

install
npmdsh plugin --profile web add dsh-vibe-math
Category
Skills & Agents
Platform
DSH Plugin
Author
chongcyrus
Distribution
Plugin

vibe-mathematics Key Features

Multi-agent math solvingFour solving architecturesResume from checkpointHuman interventionCross-validation
Listed on dsh-plugin.org

vibe-mathematics Repo Summary

What Does It Do?

Vibe Mathematics is a DSH plugin for DeepSeek Harness that provides multi-agent mathematical problem solving and formal verification. It offers four agent presets (vibe-math-v1 to v4) that collaborate to automatically solve math problems and cross-verify conclusions. It addresses the issues of single-agent solvers getting stuck and conclusions lacking credibility, letting you drive the full loop of exploration, iteration, verification, and knowledge consolidation in natural language. Core capabilities include checkpoint resume, mid-way human intervention, progress reporting, and four generations of solving architectures, with v2 and v3 as the main recommendations and v4 as an experimental resident self-organizing research architecture.

Core Features

  • Multi-agent collaboration: Explorer breaks down directions, Solver iterates, Verifier independently reviews and debates, with conclusions written to Verified/ after cross-validation.
  • Four architecture options: v1 classic pipeline (deprecated), v2 probability-driven JSON data layer, v3 paper-style md with planning agent and method library, v4 resident self-organizing research, selectable as needed.
  • Checkpoint resume and human intervention: full state persisted to disk, supports resume, and manual mode pauses at dispatch/verdict/promotion for human decisions.
  • Knowledge consolidation: propositions, methods, and subproblems stored in Propos/, Methods/, Subproblems/ libraries for reuse.
  • Natural language driven: describe the problem in natural language to start the solving process, no coding required.

How to Use This Plugin?

After installing the plugin in DSH, four agent presets appear in the preset selector; choose vibe-math-v2 or v3 (recommended) to start. Input a math problem and the plugin automatically orchestrates multi-agent solving and verification; to adjust behavior, modify parameters like probability thresholds or parallelism in the configuration, or use manual mode to intervene at key decision points. v4 is experimental, suitable for exploring resident self-organizing research scenarios.

This page is an independent rewrite of the plugin's official README — for authoritative documentation and the latest changes, refer to the source: chongcyrus/vibe-mathematics. The plugin is third-party code that runs on your machine once installed; inclusion does not imply endorsement — please review the source before installing.

Install, update, and uninstall this plugin in the Plugin Market of DeepSeek Harness's DSH Plugin Hub

DSH Plugin FAQ