Search

Install and manage OpenSocrates for Antigravity, Claude Code, Cowork, Codex, Cursor, Grok Build, and OpenCode.

Run agent fan-outs across two AI subscriptions at once: pick each subagent's model by complexity, length and token efficiency against live usage limits, spend Codex before Claude, and always grade work on the vendor that did not produce it.

在独立 DSH Session 中按计划执行编码任务，支持 Web 设置页与 Agent 双入口管理。

Dropstone SDK — the official TypeScript SDK for the Dropstone agent runtime. Programmatic access to the agent server, with one cross-surface memory (Continuity) shared across CLI, chat, and SDK.

Feature-implementation algorithm for ATDD/DDD projects built on the nextjs-fastify stack architecture.

Like a Wayback Machine for your own website. One command screenshots every page into dated folders on your machine, and shows which pages changed since your last run.

Sift Q installer CLI: registers agent harnesses with the hosted Sift Q service.

Static and dynamic quality checker for DSH plugins

pi extension + skill: visual understanding via a configurable vision model (pi-registry reuse or custom responses API; default gpt-5.6-luna) when the main model cannot see images — pure TypeScript, single runtime

Give text-only pi models vision — describe images with a vision model you pick via an interactive picker, then hand off the text description to non-vision models

Eyes for text-only DeepSeek on DeepSeek Harness: image understanding via Doubao Web by default (zero cost, no API key), Antigravity IDE quota (flash/pro), Gemini API, or Cockpit proxy — model-invokable vision tool, wrapper adapters, evidence memory, and a

DeepSeek Harness 本地视觉插件：通过 LM Studio 本地视觉模型描述图片/截图，全本地、无云端依赖。提供 vision_describe 工具（Node 原生后端，可选 Python 压缩后端）。

Image-understanding provider plugin for the DeepSeek Harness (DSH) web UI: injects ['vision'] and registers the OpenAI-compatible vision adapter, the provider directory, and the understand_image tool into ctx.vision (owned by @ran-sh/dsh-vision). The tool

Vision recognition plugin for DeepSeek Harness: paste images into the composer, recognize them via GLM-4V on the host side, and inject the result into the conversation.

A web component for math input (lofcz fork)

A web component for math input (lofcz fork)

Search and reuse every prompt you have typed, across every AI coding CLI on your machine. No required dependencies.

Spec-driven loop engineering: a repeatable construct for driving agents through numbered phase specs with enforced structure and mandatory status handoffs.

Baseline Pi behavior for session naming, history, prompt editing, fullscreen navigation, and footer styling

System-prompt & tool customizer for DeepSeek Harness with a settings-panel UI: block / replace / inject prompt sections by name, and hide tools from the model catalog by exclude/include. Config lives in the dsh settings document (namespace prompt-customiz