DSH Marketplace

dsh-agency-agents

MichengAI/dsh-agency-agents

Adds a summonable domain-expert roster that runs as subagents while the parent session keeps the task and final answer.

22124TypeScriptApache-2.0Source

Install

Add dsh-agency-agents to DeepSeek Harness

Install into

via npm

Resolves a published tarball rather than cloning the repository, and installs without any extra setup. Swap `web` for your profile name if you run another one.

via GitHub · npm package

Installing from GitHub runs the project's build script, which pnpm blocks until you allowlist it — run the command once and pnpm prints the exact key to add under `allowBuilds` in ~/.dsh/profiles/web/pnpm-workspace.yaml.

What happened when we ran it

Installed cleanly when we ran it

Every command here is run in a throwaway container against a clean profile, and the result is whatever the harness recorded — not a guess from the source. Last run 11d agoon DeepSeek Harness 0.2.2-rc.1.

Show it in your README

install verified — dshmarketplace

For maintainers: the badge serves this listing's latest sandbox verdict, so a re-run updates it on its own — and it links readers to the full result here.

Due diligence

Before you install dsh-agency-agents

  • Source of record: MichengAI/dsh-agency-agents — present in the community registry that DSH's own plugin market installs from.
  • Licensed under Apache-2.0.

Security

What installing dsh-agency-agents runs

Build scripts

No build script was blocked in our run.

Source: our sandbox install on DeepSeek Harness 0.2.2-rc.1, 11d ago.

Install hooks

package.json declares no preinstall, install or postinstall script.

Source: package.json on GitHub, read at our last sync. prepare is not recorded; it runs only on a GitHub install, and shows up under build scripts when pnpm blocks it.

README mentions

shell commands (bash, exec, spawn …)

Source: a keyword match on the README — a hint about what the plugin touches, not an analysis of its code.

Not an audit. These are the record of one install and two text scans; nobody has read this plugin's code on your behalf. Plugins run with your agent's permissions.

The AI take

What it is — Pulls named domain experts or teams into a DSH session as subagents; the parent keeps the task and writes the final answer.

Who it is for — You are running a technical-design review or a content-planning pass inside DSH on the session's current model, and you need domain-specific subagent roles to split the work. Skip it if each role must use a different model: teammates inherit the parent session's model, with no per-expert model setting.

Watch out — A clean-profile install was registered by the harness. Release 1.2.26 is rejected on DeepSeek Harness 0.1.x. If Agent Team is missing or not ready, it uses ordinary subagents with at most four in flight, and it executes shell commands.

The verdict — Worth bringing in only as a tightly gated roster: one team enabled, the rest off, online updates switched off on the plugin page.

Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.

What dsh-agency-agents does

Geometric specialist nodes converge through parallel channels into one central output block.

dsh-agency-agents is a DeepSeek Harness plugin that adds a summonable roster of domain experts and expert teams to a session. DeepSeek Harness runs selected experts as subagents while the parent session keeps the task, checks their results, resolves disagreements and produces the final answer.

The plugin provides 331 built-in experts across 22 categories, five built-in teams, and custom teams with two to eight members. Experts and teams can be selected from the management panel, summoned through the chat toolbar, or referenced with @. It registers tools including list_experts, summon_expert, summon_experts, list_expert_teams, get_expert_team and summon_expert_team. Team settings are stored in the host agency-agents namespace. Custom expert and team drafts can also be proposed through propose_custom_expert and propose_expert_team.

dsh-agency-agents is intended for code review, design, product, operations and research tasks that benefit from separate specialist perspectives. It is a poor fit when a single agent is sufficient, or when independent expert model selection is required: team members normally use the parent session’s model choice. DeepSeek Harness must support the required plugin version; release 1.2.26 no longer supports DeepSeek Harness 0.1.x. The declared terminal surface matters because summoned agents run as subagents within the parent session and can therefore operate through the session’s available capabilities.

dsh-agency-agents documentation

Configuring dsh-agency-agents

DeepSeek Harness stores expert-team configuration in the agency-agents namespace. The plugin supports an updater setting: updater: false disables online updates and provides a platform-level read-only lock. The README does not specify a separate configuration file or an environment-variable interface.

The plugin can manage enabled experts, custom experts and up to 100 custom teams. A team can contain two to eight members, with member responsibilities, shared goals, constraints, delivery requirements and one to three task examples. Enabling a team can also enable the experts it requires. Disabling or deleting a team does not disable experts or remove historical sessions.

Commands and tools

DeepSeek Harness exposes these plugin tools:

  • list_experts, summon_expert and summon_experts for individual experts.
  • list_expert_teams, get_expert_team and summon_expert_team for teams.
  • propose_custom_expert and propose_expert_team for conversational creation drafts.

The plugin also adds task recommendations in the composer. This feature is disabled by default, performs local tokenisation and scoring, and uses a 250ms typing debounce. It does not call a backend model for recommendations.

How dsh-agency-agents behaves

When Agent Team support and the current session tools are available, DeepSeek Harness uses the native Agent Team flow with shared tasks and teammate messages. Otherwise, the plugin uses ordinary subagents; it can suggest enabling Agent Team support when available but does not require it. Explicit maxDepth uses ordinary subagent mode. A normal team run allows up to four experts in parallel, retains completed results after partial failure, and does not create an additional team leader.

Requirements and limits

The README describes version 1.2.26 as a breaking release that rejects DeepSeek Harness 0.1.x; the published compatibility badge lists support through 0.2.2-rc.1. Experts cannot summon further experts or add automatic analysis rounds. Cancelling stops tasks that have not started, while completed results and failure reasons are reported separately.

Written from the project's own documentation and kept in sync with it. Where the two disagree, the source is authoritative — read the README on GitHub

Same category

Alternatives to dsh-agency-agents

dsh-skill-explorer

zhu1090093659

10k

Skill center for the dsh web GUI: browse all loaded skills grouped by source, enable or disable model invocation, create new skills, and delete into a recoverable trash.

Installed cleanly when we ran it

npm packageTypeScript2d ago

AI review

dsh-skill-explorer

What it is — A DSH Web GUI page that lists loaded skills by origin, with invocation toggles, creation, and a trash you can restore from.

Who it is for — You run agents in the DSH Web GUI with skills piled in from several origins, and the model keeps hitting ones you wanted left idle. Anyone who only edits skill files in a terminal and never opens the Web GUI has no use for this panel.

Watch out — On a fresh profile it installed and the harness registered it. No sharp edges showed up in that run.

The verdict — A single setup pass on invocation flags does not earn a plugin slot from me; I add it only when I keep turning skills on and off.

Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.Read the source
1Details

Answers a question with a one-page visual HTML explainer: the model writes a short Markdown draft and the bundled `am` CLI builds the page, with flow, sequence, tree, timeline and ER diagrams, tables and callouts.

Installed cleanly when we ran it

GitHub sourceJavaScripttoday

AI review

answer-me-with-html

What it is — Questions come back as one HTML page: the model stops at a short Markdown draft, and the bundled am CLI draws layout plus flow, sequence, tree, timeline and ER diagrams, tables and callouts.

Who it is for — You are in DeepSeek Harness unpacking a handshake, a repo module map, or Redis versus Memcached, and a terminal transcript cannot show the structure. Skip it if the agent only patches code and runs tests, or if answers already land in a docs repo you maintain—the extra file has no place in that path.

Watch out — In a fresh profile it installed and the harness registered it. Node.js 20 or newer is required, and the bundled am CLI is invoked through the shell. Their Claude Sonnet 5.5 numbers on a plain Claude Code setup: model-written tokens about 4,893 versus 612, 31 s versus 12 s, bill down only about 15% because each turn still reads the system prompt, the question and the thread.

The verdict — I would enable it on a profile reserved for explainers I will send or reopen, and leave it off sessions that die in the terminal.

Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.Read the source
1Details

skillhub

iflytek

5.3k

Browse and search a SkillHub registry and manage local skill installation, updates, and removal inside DeepSeek Harness. Targets DSH 0.2.0-rc.2 and 0.2.1-rc.2; not yet compatible with 0.2.1-rc.3.

No one-line install — this plugin lives inside a larger repository and publishes no npm package.

GitHub sourceJava25d ago

AI review

skillhub

What it is — A DeepSeek Harness plugin that queries a SkillHub registry and installs, updates, or removes skills on the local machine.

Who it is for — You do agent work inside DeepSeek Harness while the skills your team actually reuses already live in a SkillHub you can reach, and you are still jumping out to fetch packages. If skills move as git repos or shared folders and your day is not spent in DSH, skip this plugin.

Watch out — It is not published as a one-command package; you have to wire it from a repository subdirectory. It runs shell commands and expects an API key or token. It targets DSH 0.2.0-rc.2 and 0.2.1-rc.2 and is not compatible with 0.2.1-rc.3.

The verdict — This is not a reason to introduce SkillHub; I would only attach it after a registry is already how skills get distributed.

Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.Read the source
2Source

dashi-taskboard

chuspeeism

3.4k

现代化可灵活嵌入的任务面板,支持 Codex、DeepSeek Harness

Installed cleanly when we ran it

GitHub sourceJavaScript19d ago

AI review

dashi-taskboard

What it is — A local-first issue board whose HTTP API feeds both a browser UI and taskctl, with a CDP path into Codex.

Who it is for — You are cutting slices in the Codex desktop app and need the agent to inspect issues and move their status on a machine-local board. If DeepSeek Harness is your only surface and you expect a matching in-harness panel after install, skip it—the writeup never describes that UI, only Codex injection and desktop apps.

Watch out — On a clean profile it installed and the harness registered it. It runs shell commands and wants an API key or token; the service needs Node.js 22.5 or newer. The Codex launcher or injector has to stay running while the panel is used, and desktop builds add Rust 1.88 plus Xcode or Visual Studio Build Tools.

The verdict — I would keep it for the Skill’s last gate—done only after I accept—not because it registered.

Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.Read the source
2Source

vox-director

Alisa0808

2.5k

Turn one topic into a finished Vox-style paper-collage explainer/ad video — automated end to end on Atlas Cloud + ffmpeg. An agent skill.

Installed cleanly when we ran it

GitHub sourcePython1mo ago

AI review

vox-director

What it is — An agent skill that takes one topic through beat maps, collage posters, motion, voice-over, music and burned-in captions and writes a Vox paper-collage mp4 .

Who it is for — You are cutting explainer or ad shorts inside Claude Code or Codex from a topic line, a talking-head, or a product still, and you lack a skill that owns that collage path. Skip it if you only need to swap narration or BGM on an edit that already exists; this is not a detachable TTS plugin and it re-enters through beats.json .

Watch out — It calls Atlas Cloud for image, i2v , TTS and music, then shells out to local ffmpeg.You approve the beat map and pick a look by eye before the rest runs. Sandbox: it installed on a fresh profile and the harness registered it.

The verdict — I leave it registered and judge from the first posters; thin paper texture and I pull it rather than debug the video stage.

Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.Read the source
1Source
1.9k

Daily radar for new AI terms: names of new models, products and repos, caught on X, Hugging Face, GitHub and sitemaps before keyword tools list them.

Installed cleanly when we ran it

GitHub sourcePython20d ago

AI review

new-term-radar

What it is — Collects new AI model, product and repo names from X, Hugging Face, GitHub and sitemaps, ahead of keyword tools.

Who it is for — You are writing pages for an AI-tool site or picking the day's newsletter lead, and you need names that just surfaced. If your DeepSeek Harness work is code and agent wiring, this list has no downstream.

Watch out — It installed on a fresh profile and DeepSeek Harness registered it. X and new-site signals need an AIsa key and bill per run; Hugging Face, GitHub and sitemaps do not. It executes shell commands.

The verdict — I would keep it only as a daily report habit; a one-off name check is not worth a plugin slot.

Generated by grok-4.6, and a starting point rather than a verdict. Where it says a plugin installs or does not, that is from a real run in a clean profile — everything else is read off the repository. Trust the source over this.Read the source
1Source