Multi-candidate mode for DeepSeek Harness (test-time scaling): floating panel config, the agent auto-runs parallel candidates plus verifier selection, scoped to the current session.
Install
# from npm (prebuilt)
dsh plugin --profile web add dsh-multi-candidate
# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)
dsh plugin --profile web add github:WintryGrass/dsh-multi-candidate
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.
多候选模式插件(test-time scaling):让 DeepSeek Harness 对同一任务并行生成多个独立候选方案,再用验证者选优——用更多推理算力换取成功率("跑 N 遍交最好的一份")。
功能
- 🐋 悬浮球面板:可自由拖动(位置记忆),点击开/关面板
- 纯配置、零操作:勾选启用后,直接在输入框正常说任务即可,模型自动按多候选流水线执行——无需特殊指令前缀、无需复制粘贴
- 配置项:启用开关、候选数(2-5)、第二意见复核(多声音合议)、验证标准(选填)
- 仅当前会话生效:开启时绑定当前会话,新会话默认关闭,不跨会话泄漏
- 硬接线工具
multi_candidate_run:模型显式触发入口,参数化调用
要求
- DeepSeek Harness rc.8 及以上(依赖新版
defineToolAPI:execute/output/裸属性表) - web profile(客户端面板运行于 Web GUI)
- 宿主已提供 peer 依赖:
@deepseek-ai/dsh-tools、dsh-settings、dsh-system-prompt、schemastery、cordis(均 optional,宿主自带)
安装
# 从 npm(如已发布)
dsh plugin --profile web add dsh-multi-candidate
# 从本地 tarball
dsh plugin --profile web add file:/path/to/dsh-multi-candidate-1.1.0.tgz
# 从本地目录(开发)
dsh plugin --profile web add file:/path/to/dsh-multi-candidate
安装后重启 DSH,Web GUI 右下角出现 🐋 按钮。
使用
- 点右下角 🐋 打开面板
- 勾选「启用多候选模式(仅当前会话)」,可选候选数(默认 3)、第二意见、验证标准
- 在主输入框正常输入任务(如"帮我写一个 Python 猜数字游戏")
- 模型自动执行:并行 N 候选 → 验证者选优 →(可选第二意见)→ 汇报
执行时自动尊重当前会话模式(router / J-space / Anchored 等)。
工作原理
🐋 面板(客户端)→ settingsScope.set() → 服务端 settings 持久化
↓
systemPrompt.section 动态注入(仅对绑定会话)→ 模型上下文自动带配置
↓
任务正常输入 → 模型按配置调用 workflow 工具执行「并行候选→验证选优」
给开发者的注意
file:安装是复制模式:修改源码后需remove + add重装,或手动同步到 profile 的node_modules(pnpm 缓存不自动刷新)- 服务端改动必须重启 DSH;仅客户端改动刷新页面即可
- 多候选的最终执行依赖 agent 层的
workflow工具(服务端工具无法直接编排子代理,这是架构边界)
License
MIT
Links
More in this category
Q00/ouroboros#integrations/dsh-plugin★ 6183
Config-only bundle that mounts Ouroboros through the DSH MCP client, exposing 36 interview, Seed, execution, evaluation, and evolution workflow tools in DSH.
loopx-project/loopx#dsh-loopx-plugin★ 6161
LoopX, a provider-neutral, local-first state kernel and control plane for long-horizon agents: keeps Goal, Todo, gate, evidence, quota, recovery, and handoff state above DeepSeek Harness, while the plugin bootstraps the CLI and skills, admits bounded same-session continuation, and adds a loopback GoalBar for the exact bound loop.
chuspeeism/dashi-taskboard#deepseek-harness★ 3289
Embeds the active installed Codex Taskboard runtime in the DeepSeek Harness sidebar, using its launcher runtime descriptor instead of a fixed port.
NanmiCoder/dsh-agent-teams★ 1937
AgentTeams multi-agent teams.
EthanYoQ/AI-Novel-Writer#dsh-ai-novel-writer★ 1313
Installs a dedicated AI novel-writing preset and workbench: revisioned local project assets, a compact side drawer, and native approval-gated single-file changes.
tong-io/tongflow#dsh-tongflow★ 1037
TongFlow film-crew studio for image, voice, music and video production: the agent writes per-asset TongFlow workflow files (.tongflow.json) that run through TongFlow plugins, with an embedded workflow canvas, a shot/character/take project layout and a manga-drama template; sessions starting with @tongflow open the Studio view.
Community comments
Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.