DeepSeek Harness Plugin Hub

Publish and manage complete Harness Profiles. Discover Plugins for your next setup.

Explore

PluginsPresetsDocsNews

Community

Publish a pluginContactReport an issue

Resources

Plugin Hub on GitHubDeepSeek HarnessSystem statusPrivacy notice
© 2026 DeepSeek Harness Plugin HubPowered byPaxTech

Independent and unofficial. Not affiliated with, authorized by, or endorsed by DeepSeek.

Community Feature Requests — Plugin Hub News
DeepSeek Harness Plugin Hub
ProfilesPluginsCategoriesNewsDocsSign inManage Profiles
ProfilesPluginsCategoriesNewsDocsSign in

COMMUNITY CHALLENGES

Find your next challenge.

Real problems from the community. Find one that sparks an idea, explore the discussion, and build something people need.

Showing 121–140 of 255 community requestsUpdated hourly from community discussions
All updatesDSH ReleasesFeature Requests
All3 days7 days14 days
TopNew
OpenShippedSolved by pluginWon't fixAll statuses
#Problems worth solvingCommunity demand
121
HARNESS
2

Resume Sessions Across Restarts

Expose session resumption through the DSH SDK and preserve the active conversation and session state when DSH is restarted from the web interface.

Explore challenge ↗

The problem

Expose session resumption through the DSH SDK and preserve the active conversation and session state when DSH is restarted from the web interface.

Possibly addressed by:Session Resilience · Restart Autoresume · Session Conductor

Start with these discussions

  • 请问dsh-sdk啥时候会添加resume,目前重启后历史会话不可重复问答github · ▲1 · 1 comments
  • web 端使用，让AGENT自己重启DSH，重启后会丢失本轮会话内容github · ▲1 · 0 comments
Last active Sep 30, 2026
2 authors2 posts · 2 reactions1 comments
122
HARNESS
3

Add session-scoped filesystem workspaces

Give each session an isolated working directory and allow filesystem operations outside agent turns to target an explicit session scope. Propagate session identity through filesystem APIs and endpoints so shared DSH processes can safely serve multiple users.

Explore challenge ↗

The problem

Give each session an isolated working directory and allow filesystem operations outside agent turns to target an explicit session scope. Propagate session identity through filesystem APIs and endpoints so shared DSH processes can safely serve multiple users.

Possibly addressed by:Session Scope · Session Workspace · Multi Workspace

Start with these discussions

  • Feature request: session scope for `fs`github · ▲3 · 1 comments
  • 同一个工作区内的会话为啥不做个目录隔离？github · ▲1 · 5 comments
  • Session-scoped cwd: AgentLoop.create accepts one, the SDK wire has no way to send itgithub · ▲1 · 0 comments
123
PLUGIN
4

Add configurable developer-role compatibility

Make developer-role support configurable per model, defaulting to disabled for non-native OpenAI endpoints, and expose the setting in model configuration. This lets compatible providers opt in while preserving system roles for gateways that reject developer messages.

Explore challenge ↗

The problem

Make developer-role support configurable per model, defaulting to disabled for non-native OpenAI endpoints, and expose the setting in model configuration. This lets compatible providers opt in while preserving system roles for gateways that reject developer messages.

Possibly addressed by:Llm Openai Completions · Gateway Config · Custom Provider Reasoning

Start with these discussions

  • [BUG] llm-pi-ai 应支持配置 compat.supportsDeveloperRolegithub · ▲11 · 5 comments
  • [bug or limits] bug 400: {"message":"Unexpected message role.","type":"BadRequestError","param":null,"code":400} while try reasoning with llm-pi-aigithub · ▲2 · 3 comments
  • 「续 #3379」建议：supportsDeveloperRole 默认改为 false，并在模型配置 UI 增加可见开关
124
HARNESS
2

Detect stalled and idle Goal-mode continuation

Make Goal mode detect repeated no-progress rounds and waiting or idle states, pausing or stopping continuation when background work is running or no meaningful task remains. This prevents unnecessary calls, injected turns, and token consumption.

Explore challenge ↗

The problem

Make Goal mode detect repeated no-progress rounds and waiting or idle states, pausing or stopping continuation when background work is running or no meaningful task remains. This prevents unnecessary calls, injected turns, and token consumption.

Possibly addressed by:Goal Auto · Goal Pause Guard · Goal Restart

Start with these discussions

  • Goal模式空转，亟需优化github · ▲2 · 1 comments
  • goal 续行缺少"无进展"与"等待中"判断：0.1.6-alpha.2 仍复现github · ▲1 · 4 comments
Last active Sep 19, 2026
2 authors2 posts · 3 reactions5 comments
125
HARNESS
2

Show and navigate subagent model assignments

Make each subagent’s model visible and improve consistent identification, overview, and navigation when many subagents are active.

Explore challenge ↗

The problem

Make each subagent’s model visible and improve consistent identification, overview, and navigation when many subagents are active.

Possibly addressed by:Subagent Model Visibility · Subagent Model Router · Subagent Model Picker

Start with these discussions

  • 关于子智能体的几个小问题github · ▲1 · 1 comments
  • v0.2.0-rc.1为什么没有解决不显示子智能体使用的是什么模型的问题github · ▲1 · 0 comments
Last active Sep 28, 2026
2 authors2 posts · 2 reactions1 comments
126
HARNESS
3

Make fs-ext optional for installation

Allow DSH to install and run without a native C++ toolchain by skipping the POSIX-only fs-ext dependency when unnecessary, using prebuilt alternatives, or providing clear prerequisite guidance. Windows installations should not require Visual Studio build tools solely to compile fs-ext.

Explore challenge ↗

The problem

Allow DSH to install and run without a native C++ toolchain by skipping the POSIX-only fs-ext dependency when unnecessary, using prebuilt alternatives, or providing clear prerequisite guidance. Windows installations should not require Visual Studio build tools solely to compile fs-ext.

Possibly addressed by:Windows Native · Windows Bash · Minimal Msys2

Start with these discussions

  • Windows 更新后源码运行失败：fs-ext@2.1.1 在 node-gyp 阶段无法完成安装 （Windows source install fails at fs-ext@2.1.1 / node-gyp after recent update）github · ▲4 · 4 comments
  • @deepseek-ai/dsh@0.1.3-alpha.2 cannot be installed without a C++ toolchain: new hard dependency fs-ext (node-gyp, no prebuilt binaries)github · ▲1 · 2 comments
127
HARNESS
2

Select Model and Effort Per Session

Allow each session or launch to choose its own model and effort level without changing the deployment-wide default. Propagate these settings through session creation and agent resolution.

Explore challenge ↗

The problem

Allow each session or launch to choose its own model and effort level without changing the deployment-wide default. Propagate these settings through session creation and agent resolution.

Possibly addressed by:Subagent Model Picker · Effort · Agent Team Model

Start with these discussions

  • Session-scoped model and effort selectiongithub · ▲4 · 4 comments
  • Session-only model selection for Agent Sessions lanesgithub · ▲2 · 1 comments
Last active Sep 19, 2026
1 authors2 posts · 6 reactions5 comments
128
PLUGIN
3

Support input modalities in gateway model listings

Add discovery support for flat and nested input modality fields in gateway model listings, including OpenRouter, while filtering unknown values and preserving unknown status when absent. This allows vision-capable models to accept image attachments instead of being treated as text-only.

Explore challenge ↗

The problem

Add discovery support for flat and nested input modality fields in gateway model listings, including OpenRouter, while filtering unknown values and preserving unknown status when absent. This allows vision-capable models to accept image attachments instead of being treated as text-only.

Possibly addressed by:Models Input Modalities · Gateway Models · Model Modality

Start with these discussions

  • feat(llm-pi-ai): read input modalities from gateway model listings (patch ready on a fork branch)github · ▲1 · 0 comments
  • feat(llm-pi-ai): read input modalities from gateway model listings (patch ready on a fork branch)github · ▲1 · 0 comments
129
HARNESS
3

Add UI editing for per-model capabilities

Provide a settings or dedicated capability editor for per-model input modalities, reasoning efforts, and compatibility fields, including model overrides. Ensure these fields are persisted and not discarded by later GUI edits, eliminating the need for hand-editing YAML.

Explore challenge ↗

The problem

Provide a settings or dedicated capability editor for per-model input modalities, reasoning efforts, and compatibility fields, including model overrides. Ensure these fields are persisted and not discarded by later GUI edits, eliminating the need for hand-editing YAML.

Possibly addressed by:Model Capabilities · Model Capability · Awesome Model Setting

Start with these discussions

  • Settings: the model capability fields the schema already supports are not editable in any UIgithub · ▲1 · 0 comments
  • Settings: the model capability fields the schema already supports are not editable in any UIgithub · ▲1 · 0 comments
  • Settings: the model capability fields the schema already supports are not editable in any UI
130
HARNESS
2

Add chat limit indicators and continuation

Provide a visible token or percentage counter, advance warnings before the context limit, and ways to continue a truncated chat by carrying its history or an automatic summary into a new chat. Also provide structured conversation export as a fallback to prevent loss of context and work.

Explore challenge ↗

The problem

Provide a visible token or percentage counter, advance warnings before the context limit, and ways to continue a truncated chat by carrying its history or an automatic summary into a new chat. Also provide structured conversation export as a fallback to prevent loss of context and work.

Possibly addressed by:Chat History · Chat Offscreen · Chat History

Start with these discussions

  • One feature request to change to my life.github · ▲1 · 0 comments
  • chat reaches its maximum length, the conversation ends without any warninggithub · ▲1 · 0 comments
Last active Oct 2, 2026
2 authors
131
HARNESS
1hackernews1

Add Customizable Keyboard Shortcuts

Provide discoverable, customizable keyboard shortcuts for navigation, chat and input focus, commands, permissions, prompt handling, copying, and stopping interactions, reducing reliance on the mouse.

Explore challenge ↗

The problem

Provide discoverable, customizable keyboard shortcuts for navigation, chat and input focus, commands, permissions, prompt handling, copying, and stopping interactions, reducing reliance on the mouse.

Possibly addressed by:Keyboard Manager · Shortcuts · Conversation Shortcuts

Start with these discussions

  • 快捷键体验优化建议github · ▲1 · 1 comments
  • Comment on: Pi 1.0hackernews · ▲0 · 0 comments
Last active Oct 2, 2026
2 authors2 posts · 1 reactions1 comments
132
PLUGIN
2

Map pi-ai reasoning tokens

Map the pi-ai provider’s reasoning-token count to `TokenUsage.reasoningTokens` while preserving output-token counts. Omit the field when the count is zero so reasoning and answer usage can be separated consistently across transcripts, context meters, and per-turn accounting.

Explore challenge ↗

The problem

Map the pi-ai provider’s reasoning-token count to `TokenUsage.reasoningTokens` while preserving output-token counts. Omit the field when the count is zero so reasoning and answer usage can be separated consistently across transcripts, context meters, and per-turn accounting.

Possibly addressed by:Plugin Reasoning Efforts · Thinking Token Stat · Custom Provider Reasoning

Start with these discussions

  • llm-pi-ai drops the reasoning-token breakdown pi-ai already reportsgithub · ▲2 · 2 comments
  • Map pi-ai's reasoning usage into `reasoningTokens`github · ▲1 · 2 comments
Last active Sep 18, 2026
133
HARNESS
2

Warn and guide users about image inputs

Clearly indicate when a model is treated as text-only or lacks image-input capability, and warn when images are blocked or dropped. Guide users to the per-model modality settings and make those settings easy to discover.

Explore challenge ↗

The problem

Clearly indicate when a model is treated as text-only or lacks image-input capability, and warn when images are blocked or dropped. Guide users to the per-model modality settings and make those settings easy to discover.

Possibly addressed by:Image Vision · Models Input Modalities · Image Subagent

Start with these discussions

  • [Feature] 图片被拦时提示指向「切换模型」，而开关藏在折叠的「模型选项」里github · ▲1 · 0 comments
  • [UX] Models without explicit input capability are silently treated as text-only; image attachments fail with no warninggithub · ▲1 · 0 comments
Last active Oct 1, 2026
2 authors2 posts · 2 reactions
134
HARNESS
2

Virtualize long conversation rendering

Keep only recent turns mounted and represent older content with expandable folded entries or equivalent virtualization. This keeps DOM and reconciliation costs bounded so long-running conversations remain responsive.

Explore challenge ↗

The problem

Keep only recent turns mounted and represent older content with expandable folded entries or equivalent virtualization. This keeps DOM and reconciliation costs bounded so long-running conversations remain responsive.

Possibly addressed by:Chat Offscreen · Chat Window Fold · Conversation Rollback

Start with these discussions

  • 长任务对话后页面卡死无法动弹github · ▲1 · 2 comments
  • [性能] Chat 折叠只隐藏不卸载：把渲染窗口收缩为「最新 N 轮 + 1 个折叠行」（含实现与实测）github · ▲1 · 0 comments
Last active Sep 23, 2026
2 authors2 posts · 2 reactions2 comments
135
HARNESS
2

Enforce global subagent delegation budgets

Add hierarchical delegation limits, including effective concurrency caps, per-parent budgets, and whole-tree quotas, with clear rejection errors. Batch or consolidate settlement notifications and parent wakeups to prevent uncontrolled fan-out and repeated full-context runs.

Explore challenge ↗

The problem

Add hierarchical delegation limits, including effective concurrency caps, per-parent budgets, and whole-tree quotas, with clear rejection errors. Batch or consolidate settlement notifications and parent wakeups to prevent uncontrolled fan-out and repeated full-context runs.

Possibly addressed by:Subagent Concurrency Limit · Configurable Subagents · Subagent Profile

Start with these discussions

  • 子代理只有深度上限、没有广度/预算上限：一次会话产生 146 个子代理会话、3.43 亿缓存读取（占我全部历史 62%）github · ▲5 · 1 comments
  • subagent数量的上限无法有效设置github · ▲1 · 1 comments
Last active Sep 14, 2026
2 authors
136
HARNESS
2

Persist and customize sidebar widths

Persist resized left and right sidebar widths across sessions and restarts, including the context and file-preview panel, and allow users to configure suitable default or viewport-relative widths.

Explore challenge ↗

The problem

Persist resized left and right sidebar widths across sessions and restarts, including the context and file-preview panel, and allow users to configure suitable default or viewport-relative widths.

Possibly addressed by:Rightbar Default Width · Sidebar Width · Settings Size

Start with these discussions

  • [功能建议] 右栏（上下文面板）默认宽度可自定义并持久化github · ▲1 · 0 comments
  • [Feature Request] sidebars support persistent widthgithub · ▲1 · 0 comments
Last active Sep 29, 2026
2 authors2 posts · 2 reactions0 comments
137
HARNESS
hackernews4

Increase Parallel Subagent Capacity

Enable DSH to run more specialized subagents and agent sessions concurrently for parallel work, improving throughput while allowing shorter, task-focused contexts.

Explore challenge ↗

The problem

Enable DSH to run more specialized subagents and agent sessions concurrently for parallel work, improving throughput while allowing shorter, task-focused contexts.

Possibly addressed by:Subagent Concurrency Limit · Parallel Agent Ade · Subagent Team

Start with these discussions

  • Comment on: GPT 6.1 Sol: Near-Astra intelligence for a fifth of the pricehackernews · ▲0 · 0 comments
  • Comment on: Sonnet 5.5hackernews · ▲0 · 0 comments
  • Comment on: DeepSeek v4.1 Flash Is Now Our Best Hacking Modelhackernews · ▲0 · 0 comments
Last active Sep 30, 2026
4 authors4 posts · 0 reactions0 comments
138
HARNESS
1hackernews1

Support Android and Termux

Enable official installation and execution of DSH on Android, including Termux, with the required portability, filesystem, runtime, and installation handling.

Explore challenge ↗

The problem

Enable official installation and execution of DSH on Android, including Termux, with the required portability, filesystem, runtime, and installation handling.

Possibly addressed by:Android Agent · Android · Plugin Android Use

Start with these discussions

  • dsh runs on Termux (Android) — with 5 small patchesgithub · ▲9 · 0 comments
  • Comment on: Ask HN: What are you working on? (September 2026)hackernews · ▲0 · 0 comments
Last active Sep 14, 2026
2 authors2 posts · 9 reactions0 comments
139
HARNESS
3

Support configurable inference tiers

Add provider-aware, user-selectable compute or service tiers with distinct cost, latency, and capability trade-offs, optional automatic selection, and end-to-end propagation through model and session configuration. Preserve defaults, persist choices, expose usage metadata, and never silently downgrade unsupported tiers.

Explore challenge ↗

The problem

Add provider-aware, user-selectable compute or service tiers with distinct cost, latency, and capability trade-offs, optional automatic selection, and end-to-end propagation through model and session configuration. Preserve defaults, persist choices, expose usage metadata, and never silently downgrade unsupported tiers.

Possibly addressed by:Model Tier · Tier Router · Better Model Provider

Start with these discussions

  • Feature request: end-to-end support for OpenAI Codex Standard/Fast service tiersgithub · ▲1 · 1 comments
  • User-selectable compute tiers with different token pricing for DeepSeek V4.1 Flashgithub · ▲1 · 0 comments
  • Proposal: configurable OpenAI-compatible service tier for llm-pi-ai
140
HARNESS
1hackernews1

Control subagent notifications

Let completed subagents send messages to the orchestrator and optionally wake it only when action is needed. Preserve message content verbatim and support controls over the requested follow-up task and effort.

Explore challenge ↗

The problem

Let completed subagents send messages to the orchestrator and optionally wake it only when action is needed. Preserve message content verbatim and support controls over the requested follow-up task and effort.

Possibly addressed by:Configurable Subagents · Subagent Profile · Wait Subagent

Start with these discussions

  • 建议：让子 Agent 按情况决定是否唤醒主 Agent，而不是每次都唤醒github · ▲1 · 1 comments
  • Comment on: Prompting Claude Opus 5.5hackernews · ▲0 · 0 comments
Last active Sep 28, 2026
2 authors
← PreviousPage 7 of 13Next →
Last active Sep 11, 2026
3 authors3 posts · 5 reactions6 comments
github · ▲2 · 2 comments
Last active Aug 29, 2026
4 authors4 posts · 17 reactions12 comments
Windows install requires the full MSVC toolchain for fs-ext, which win32 never actually calls / Windows 安装被迫为从未调用的 fs-ext 拉起整套 MSVC 工具链
github · ▲1 · 0 comments
Last active Sep 8, 2026
3 authors3 posts · 6 reactions6 comments
feat(llm-pi-ai): read input modalities from gateway model listings (patch ready on a fork branch)
github · ▲1 · 0 comments
Last active Oct 4, 2026
1 authors3 posts · 3 reactions0 comments
github · ▲1 · 0 comments
Last active Oct 4, 2026
1 authors3 posts · 3 reactions0 comments
2 posts · 2 reactions
0 comments
2 authors
2 posts · 3 reactions
4 comments
0 comments
2 posts · 6 reactions
2 comments
github · ▲1 · 0 comments
Last active Sep 14, 2026
3 authors3 posts · 3 reactions1 comments
2 posts · 1 reactions
1 comments