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.

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 61–79 of 79 community requestsUpdated hourly from community discussions
All updatesDSH ReleasesFeature Requests
All3 days7 days14 days
TopNew
OpenShippedSolved by pluginWon't fixAll statuses

Window Oct 21, 2026 – Nov 4, 2026

#Problems worth solvingCommunity demand
61▲25
HARNESS
4

Build robust provider-aware context management

Provide proactive, configurable compaction with model-visible pressure signals, bounded tool-output spilling, and repeat-call limits. Base it on actual provider limits and ensure deterministic progress with model-free fallbacks when summarization or oversized requests fail.

Explore challenge ↗

The problem

Provide proactive, configurable compaction with model-visible pressure signals, bounded tool-output spilling, and repeat-call limits. Base it on actual provider limits and ensure deterministic progress with model-free fallbacks when summarization or oversized requests fail.

Possibly addressed by:Smart Compaction · Handoff Compaction · Improved Compact

Start with these discussions

  • token 消耗过大的根因分析及改进建议github · ▲2 · 6 comments
  • [Proposal] Provider-neutral prompt token pressure + local overflow -> compact-retrygithub · ▲1 · 6 comments
  • Compaction cannot rescue an oversized session — the summarization request itself exceeds the provider request-size limit (HTTP 413)github · ▲1 · 3 comments
Last active Oct 1, 2026
0 authors0 posts · 1 reactions1 comments
62▼11
HARNESS
4

Configure per-conversation session-affinity headers

Attach stable per-conversation session identifiers, such as UUID-based OpenCode headers, to outbound inference requests. Allow provider- and model-level control over header names, formats, and omission, together with related prompt-cache settings to improve routing and cache affinity.

Explore challenge ↗

The problem

Attach stable per-conversation session identifiers, such as UUID-based OpenCode headers, to outbound inference requests. Allow provider- and model-level control over header names, formats, and omission, together with related prompt-cache settings to improve routing and cache affinity.

Possibly addressed by:Opencode Session Header · X Opencode Session · Session Header

Start with these discussions

  • Please send x-opencode-session header on API requests (OpenCode Go requires it from 09/05)github · ▲187 · 47 comments
  • Poor prompt-cache hit rate on z.ai GLM Coding Plan: prefix is stable, hits still collapse (evidence + proposal to expose pi-ai session-affinity switches for the zai routes)
63
HARNESS
6

Support conversation groups and workspace-less sessions

Allow users to organize conversations into user-defined groups, including chats started without a workspace or within a workspace. Request a working directory only when execution requires one.

Explore challenge ↗

The problem

Allow users to organize conversations into user-defined groups, including chats started without a workspace or within a workspace. Request a working directory only when execution requires one.

Possibly addressed by:Just Chat · Auto Workspace · Plugin No Workspace

Start with these discussions

  • 兄弟们，没办法创建不带工作区的全局对话啊，这不是问题么？github · ▲2 · 10 comments
  • support grouping sessionsgithub · ▲1 · 2 comments
  • feat(sidebar): first-class workspace-less "chat" sessions — Work/Chat affordance in the sidebar and a "no workspace" option in the pickergithub · ▲1 · 2 comments
Last active Sep 29, 2026
64▲12
HARNESS
4

Implement scoped, persistent agent memory

Provide a backend-agnostic, first-party memory plugin or service with session, agent, tenant, topic, and repository scoping; automatic capture and recall; safe lifecycle handling; provenance, deduplication, bounded injection, and tools for managing facts and lessons. Support importing existing memories and preserving history safely across migrations.

Explore challenge ↗

The problem

Provide a backend-agnostic, first-party memory plugin or service with session, agent, tenant, topic, and repository scoping; automatic capture and recall; safe lifecycle handling; provenance, deduplication, bounded injection, and tools for managing facts and lessons. Support importing existing memories and preserving history safely across migrations.

Possibly addressed by:Agent Memory · Plugin Memory · Hybrid Memory

Start with these discussions

  • dsh-plugin: Native OpenViking memory for DeepSeek Harness (automatic session capture + memory tools + profile injection)github · ▲2 · 5 comments
  • 给 #14 补点料：两个现成记忆实现里踩过的坑
65▼7
HARNESS
33

Upgrade pi-ai for current provider models

Upgrade pi-ai to at least 0.87.1 to expose current OpenCode Go Flash models, including V4.1 Flash, along with current DeepSeek and Mistral support and required session headers. Preserve the streaming-performance patch and improve configuration guidance as needed.

Explore challenge ↗

The problem

Upgrade pi-ai to at least 0.87.1 to expose current OpenCode Go Flash models, including V4.1 Flash, along with current DeepSeek and Mistral support and required session headers. Preserve the streaming-performance patch and improve configuration guidance as needed.

Possibly addressed by:Opencode Go Path · Opencode Go · Opencode Go

Start with these discussions

  • opencode-go provider returns stale model list — deepseek-flash (V4.1 Flash) missing & custom provider fails with MissingSessionID
66
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 · 2 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
67▲2
HARNESS
3

Scope MCP Configuration to Workspaces

Allow each project or workspace to define and automatically use its own isolated MCP servers, tools, and settings across supported session interfaces without affecting other workspaces.

Explore challenge ↗

The problem

Allow each project or workspace to define and automatically use its own isolated MCP servers, tools, and settings across supported session interfaces without affecting other workspaces.

Possibly addressed by:Workspace Mcp · Mcp Workspace Scope · Workspace Mcp

Start with these discussions

  • Workspace-scoped MCP registration — or is CLI + workspace skill the intended pattern?github · ▲2 · 2 comments
  • 是否支持项目级的MCP配置？ 比如codex 是在项目下的 config.toml，claude是在项目下的 mcp.jsongithub · ▲2 · 0 comments
  • [Feature] 支持工作区级 MCP 配置github · ▲1 · 0 comments
68
HARNESS
12

Support extensible plugin session events

Provide a stable plugin API for persisting custom, safely ignorable session events and generic trajectory records that older readers can skip, recover, and render. Include pre-write observation hooks with documented validation, ordering, failure, and persistence behavior.

Explore challenge ↗

The problem

Provide a stable plugin API for persisting custom, safely ignorable session events and generic trajectory records that older readers can skip, recover, and render. Include pre-write observation hooks with documented validation, ordering, failure, and persistence behavior.

Possibly addressed by:Session Note · Session Guard · Session Nexus

Start with these discussions

  • [Ecosystem] Please open the deferred session-event registration surface — 12 threads, 9+ consumers, and 0.1.2-alpha left writers no path at allgithub · ▲4 · 4 comments
  • `Session.append` cannot set `ignorable`, so a plugin's own event type makes its sessions unloadablegithub · ▲3 · 2 comments
  • [Proposal] Session event registration for out-of-tree plugins — unknown event types refuse resume (dsh-click/observed|action crashes rc.6/rc.7)
69
HARNESS
1hackernews51

Require confirmation for risky actions

Require explicit approval for destructive, costly, irreversible, or production-impacting actions, including when full-access mode normally disables approval. Support configurable boundaries such as protected paths, spending limits, and destructive-only confirmation.

Explore challenge ↗

The problem

Require explicit approval for destructive, costly, irreversible, or production-impacting actions, including when full-access mode normally disables approval. Support configurable boundaries such as protected paths, spending limits, and destructive-only confirmation.

Possibly addressed by:Agent Guard Dsh · Agent Approval · Always Require Tools Approval

Start with these discussions

70▼14
HARNESS
2hackernews1

Notify on agent state changes

Add opt-in notifications for agent waits, approvals, escalations, errors, and task or round completion through configurable channels, including system, browser, and audio alerts. Expose these state transitions as public events for plugins and presets, with controls for hidden versus active tabs and optional notification actions.

Explore challenge ↗

The problem

Add opt-in notifications for agent waits, approvals, escalations, errors, and task or round completion through configurable channels, including system, browser, and audio alerts. Expose these state transitions as public events for plugins and presets, with controls for hidden versus active tabs and optional notification actions.

Possibly addressed by:Chime Alerts · Agent Notify · Notify

Start with these discussions

  • Feature request — automatic notifications when an agent pauses for user input or completes a long-running flowgithub · ▲2 · 6 comments
71
PLUGIN
2

Add persistent whole-interface zoom shortcuts

Support keyboard shortcuts and reset controls for scaling the entire client—including panels, sidebars, overlays, controls, and conversation content—with the selected scale persisted across restarts.

Explore challenge ↗

The problem

Support keyboard shortcuts and reset controls for scaling the entire client—including panels, sidebars, overlays, controls, and conversation content—with the selected scale persisted across restarts.

Possibly addressed by:Zoom · Plugin Scaling · Page Zoom

Start with these discussions

  • [Feature] Zoom the whole client UI with ⌘+ / ⌘-github · ▲5 · 3 comments
  • dsh-zoom — Cmd/Ctrl +, -, 0 to scale the whole interface (not just conversation text)github · ▲1 · 0 comments
Last active Oct 2, 2026
0 authors0 posts · 0 reactions1 comments
72
PLUGIN
21

Create a configurable global shortcut registry

Provide a centralized, customizable shortcut system where plugins can register bindings and callbacks, while users can view, edit, and resolve conflicts among shortcuts. Expose stable public APIs for shortcut integration.

Explore challenge ↗

The problem

Provide a centralized, customizable shortcut system where plugins can register bindings and callbacks, while users can view, edit, and resolve conflicts among shortcuts. Expose stable public APIs for shortcut integration.

Possibly addressed by:Keyboard Manager · Hotkeys Platform · Shortcuts

Start with these discussions

  • DeepSeek Harness 的快捷键设置应该提供第三方插件的快捷键注册能力。x · ▲1 · 1 comments
  • 全局键盘快捷键注册插件, 希望有一个插件可以处理其他全部插件提供的快捷键github · ▲1 · 1 comments
73
HARNESS
4hackernews1

Support Workspace Relocation and Rebinding

Provide a supported workflow to rebind workspaces after folder moves or renames while preserving sessions and updating related paths, configuration, history, metadata, and caches. Safely isolate corrupt or mismatched sessions so relocation does not prevent startup.

Explore challenge ↗

The problem

Provide a supported workflow to rebind workspaces after folder moves or renames while preserving sessions and updating related paths, configuration, history, metadata, and caches. Safely isolate corrupt or mismatched sessions so relocation does not prevent startup.

Possibly addressed by:Session Workspace · Workspace Mover · Session Mover

Start with these discussions

  • Renaming a workspace folder makes its stored sessions fail the identity check, with no supported repairgithub · ▲2 · 2 comments
  • [Bug] 项目目录迁移后 dsh 启动崩溃（corrupt session log），希望支持会话重新归属
74▼9
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 · 5 comments
  • Session-only model selection for Agent Sessions lanesgithub · ▲2 · 1 comments
Last active Sep 19, 2026
0 authors0 posts · 0 reactions1 comments
75
PLUGIN
2hackernews42

Provide Remote-Control Clients for DSH

Provide polished desktop and mobile clients for connecting to remote DSH servers, monitoring and steering agents, reviewing work, taking over when needed, and returning conveniently to terminal workflows.

Explore challenge ↗

The problem

Provide polished desktop and mobile clients for connecting to remote DSH servers, monitoring and steering agents, reviewing work, taking over when needed, and returning conveniently to terminal workflows.

Possibly addressed by:Mobile Remote · Phone Remote · Remote X

Start with these discussions

  • 桌面端和手机端远程连接服务器上的dsh-web ?github · ▲1 · 2 comments
76▼60
HARNESS
2

Refresh MCP tools after reconnect

Allow long-lived sessions to re-fetch and register MCP tools after a connection failure or successful reconnect, so recovered servers can be used without restarting the session.

Explore challenge ↗

The problem

Allow long-lived sessions to re-fetch and register MCP tools after a connection failure or successful reconnect, so recovered servers can be used without restarting the session.

Possibly addressed by:Live Mcp · Mcp Skill Control · Mcp Skill Control

Start with these discussions

  • mcp-client: stateful MCP server restart leaves every tool call failing until DSH restartsgithub · ▲2 · 5 comments
  • Headless session env: are DSH_-prefixed variables intentionally dropped before reaching the session?github · ▲1 · 1 comments
Last active Oct 15, 2026
0
77
HARNESS
1hackernews2

Enforce user and project instructions

Make the agent reliably follow the user's intent and applicable project instruction files such as AGENTS.md, while respecting higher-priority system, developer, and direct user instructions when conflicts occur.

Explore challenge ↗

The problem

Make the agent reliably follow the user's intent and applicable project instruction files such as AGENTS.md, while respecting higher-priority system, developer, and direct user instructions when conflicts occur.

Possibly addressed by:Agent Instructions As Prompt · Agent Instructions Plus · Custom Instructions

Start with these discussions

  • 所有注入的 AGENTS.md 指令都被"仅供参考"式前言弱化，建议改为约束性表述github · ▲2 · 1 comments
  • Comment on: Claude Code now reads AGENTS.md if there is no Claude.mdhackernews · ▲0 · 0 comments
78
HARNESS
3

Replay Session History on ACP Load

Add ACP session/load support and replay persisted session updates when loading or resuming a session. This lets clients rebuild the full conversation transcript after restart instead of receiving only the resumed state and future updates.

Explore challenge ↗

The problem

Add ACP session/load support and replay persisted session updates when loading or resuming a session. This lets clients rebuild the full conversation transcript after restart instead of receiving only the resumed state and future updates.

Possibly addressed by:Acp Replay · Autoload History · Shl Session History

Start with these discussions

  • [dsh-acp] ACP clients cannot rebuild a session transcript: no session/load, and session/resume does not replay historygithub · ▲3 · 5 comments
  • [ACP] session/resume restores the agent but does not replay history; loadSession is missing, so ACP clients show a blank transcriptgithub · ▲2 · 4 comments
  • ACP: @deepseek-ai/dsh-acp does not implement session/load, so clients cannot restore conversation history
79
PLUGIN
6hackernews4

Track model usage and enforce spending budgets

Record durable, accurately attributed usage and cost events for foreground, auxiliary, tool, compaction, retry, and subagent calls. Provide session totals, warnings, and configurable spending circuit breakers.

Explore challenge ↗

The problem

Record durable, accurately attributed usage and cost events for foreground, auxiliary, tool, compaction, retry, and subagent calls. Provide session totals, warnings, and configurable spending circuit breakers.

Possibly addressed by:Usage Cost · Token Usage Cost · Token Use

Start with these discussions

  • `tokenUsage` projection never folds `compaction/summary.usage`, and replaces rather than adds across a retried stepgithub · ▲4 · 28 comments
  • [Bug] Harness 界面 Token 统计严重偏低，实际账单 12 分钟扣费 50+ 元，请核查并优化github · ▲4 · 2 comments
  • [反馈] web_search 的两个成本相关点：调用 token 未计入会话用量、单次调用成本偏高
← PreviousPage 4 of 4Next →
github · ▲1 · 2 comments
  • bug(llm-pi-ai): openai-responses 路由无法关闭 session_id 请求头github · ▲1 · 2 comments
    • Last active Sep 10, 2026
    0 authors0 posts · 0 reactions2 comments
    0 authors0 posts · 0 reactions2 comments
    github · ▲1 · 6 comments
  • 是否可以支持多租户记忆隔离，提供详细的api文档github · ▲1 · 3 comments
    • Last active Sep 25, 2026
    0 authors0 posts · 0 reactions2 comments
    github · ▲3 · 3 comments
  • Upgrade @earendil-works/pi-ai to ^0.87.1 for the new model catalog — and fix model discovery to ask live endpoints (incl. defaults)github · ▲2 · 6 comments
  • 内置 @earendil-works/pi-ai 0.85.1 版本过低：OpenCode Go 无法添加最新模型，请求也因缺少 x-opencode-session 被拒github · ▲1 · 1 comments
    • Last active Oct 20, 2026
    0 authors0 posts · 0 reactions1 comments
    github · ▲1 · 0 comments
    Last active Sep 14, 2026
    0 authors0 posts · 0 reactions1 comments
    Last active Oct 4, 2026
    0 authors0 posts · 0 reactions1 comments
    github · ▲2 · 14 comments
    Last active Oct 7, 2026
    0 authors0 posts · 0 reactions1 comments
    [Feature] danger-full-access + approval=never 组合下，不可逆删除缺少最终确认（附真实事故）
    github · ▲1 · 1 comments
  • Comment on: Claude Opus 5.5hackernews · ▲0 · 0 comments
  • Comment on: GPT-6 Sol and Lunahackernews · ▲0 · 0 comments
    • Last active Oct 2, 2026
    0 authors0 posts · 0 reactions1 comments
    浏览器声音+notification 提示完成一轮
    github · ▲2 · 2 comments
  • Comment on: Show HN: Share your AI Setup, Learn from othershackernews · ▲0 · 0 comments
    • Last active Sep 23, 2026
    0 authors0 posts · 0 reactions1 comments
  • 我们希望可以进行快捷键的自定义设置github · ▲1 · 0 comments
    • Last active Sep 25, 2026
    0 authors0 posts · 0 reactions1 comments
    github · ▲2 · 2 comments
  • 工作区文件夹绝对位置改变后工作区重绑定功能需求github · ▲2 · 1 comments
    • Last active Sep 27, 2026
    0 authors0 posts · 0 reactions1 comments
    win,安卓客户端连接 dsh web
    github · ▲1 · 0 comments
  • Okay so I'm spoiled by ChatGPT Codex Remote Control. What compares in the OSS world?x · ▲0 · 1 comments
    • Last active Oct 2, 2026
    0 authors0 posts · 0 reactions1 comments
    authors
    0 posts · 0 reactions
    1 comments
    Comment on: AI agents lie, cheat and steal. That is putting off users
    hackernews · ▲0 · 0 comments
    Last active Sep 19, 2026
    0 authors0 posts · 0 reactions1 comments
    github · ▲1 · 0 comments
    Last active Sep 21, 2026
    0 authors0 posts · 0 reactions1 comments
    github · ▲2 · 1 comments
    Last active Sep 19, 2026
    0 authors0 posts · 0 reactions1 comments