DSH Plugins Marketplace

DSH Plugins

Plugins

/

Workflow & Automation

/

dsh-agent-team-policy

d

dsh-agent-team-policy

Manifest valid

Adds one system-prompt section that limits delegation to subagents and teammates to two cases: adversarial review of freshly produced output, and one-shot context reading that returns a summary.

hasBundlePatch

dsh-agent-team-policy

A DeepSeek Harness (dsh) plugin that decides when the agent may spawn subagents or teammates — instead of leaving the harness's "delegate in the background by default" guidance as the only rule in context.

One system-prompt section. No tools, no services, no dependencies.

Why

Two things are worth a fresh context:

  1. Adversarial review. A model reviewing its own output in the same context tends to confirm its own judgement. A reviewer that cannot see the reasoning that produced the code has a real chance of finding the counterexample.
  2. One-shot context. Searching a large codebase or reading long logs bulges the main loop. Send the reading out, get back a summary, an evidence trail and the open questions — not the raw material.

Everything else — a small edit, a decision you cannot make without seeing the source, work you already understand — belongs in the main loop.

What it injects

One section, delegation:policy, at prompt order 100:

  • after the persona (0),
  • before plan mode (500), the Agent Teams policy (600) and every tool section (1000+),
  • so it is read before any tool guidance that says "start independent delegations together".

The text allows exactly the two cases above, lists the cases that must stay local (single-file change, need the raw material to decide, no verifiable per-subtask output, already know the answer), and requires every delegation to state its goal, acceptance criteria, write scope, and whether it needs this conversation. Parallel work must not overlap write scopes; reuse an idle subagent before starting another; no padding. It also requires checking with the user before every delegation — nothing is started automatically unless the user explicitly asked for it.

Install

dsh plugin add dsh-agent-team-policy

From a local checkout or any directory containing this package:

dsh plugin add /absolute/path/to/dsh-agent-team-policy

Or install the directory as a bundle from the app's Plugins page. Nothing else to configure: the bundle patch mounts one host plugin, the plugin registers one section.

Verify

Start a new session and ask the agent to repeat its delegation rules, or take a one-file change and watch that it does not delegate. The section is visible in the assembled system prompt; it is also inherited by every subagent a session spawns, so grandchildren are constrained by the same rule.

Limits

  • Guidance, not enforcement. It cannot block a subagent or spawn_teammate call; a harness-level cap (maxDepth, toolFilter) still belongs in the tool config.
  • Appended, not replaced. @deepseek-ai/dsh-tool-subagent ships its own section that still says "background by default"; the policy text explicitly claims precedence over it.
  • complete: true presets drop it. A preset that replaces the whole system prompt (the shipped minimal preset does) keeps only its own persona.
  • The rule text is Chinese. Edit POLICY in lib/index.js to change the wording or the language; the plugin deliberately has no config schema, so there is nothing to validate and nothing to break on mount.

Releasing

See CONTRIBUTING.md: publishing uses an npm Automation token (bypasses 2FA) and never an interactive one-time password, and it always passes --registry=https://registry.npmjs.org explicitly, because the default registry on a mirror-configured machine does not accept publishes.

Compatibility

Written against dsh 0.2.2-rc.1 and only uses ctx.systemPrompt.section({ name, order, text }). Any build that exposes the systemPrompt service works.

License

MIT — see LICENSE.

Comments

Loading…

Similar plugins

dsh-eteams

Captain-led multi-agent teams that split a goal into tasks and dispatch them along an execution chain to members running in their own sub-sessions, each with its own model.

Workflow & AutomationManifest valid

★ 1

↓ 316/wk

MIT

TypeScript

dsh plugin --profile web add dsh-eteams

Group-chat multi-agent orchestration console for DSH: autonomous decision-making, task dispatch, verification and iterative optimization for long tasks; bundles its own backend, installs with `dsh plu

Workflow & AutomationManifest valid

★ 0

dsh plugin --profile web add dsh-agent-hub

Reusable subagent definitions managed from a Settings page and dispatched through a single delegate_agent tool; ships with a built-in read-only Explorer preset and a nested-delegation policy toggle.

Workflow & AutomationManifest valid

★ 0

dsh plugin --profile web add dsh-custom-subagents

by gezi-wen

Ships a skill about handing work to subagents instead of doing it inline - which tasks are worth fanning out, how many to run at once, how to send them in parallel, and how to check what they returned

Workflow & AutomationManifest valid

★ 0

MIT

Sep 20, 2026

dsh plugin --profile web add dsh-subagent-delegation

by joekytc

Run multi-agent task pipelines on DSH like a team — plan, execute, review, and deliver code through a visual kanban with provable completion.

Workflow & AutomationManifest valid

★ 6

↓ 292/wk

MIT

TypeScript

Oct 9, 2026

dsh plugin --profile web add @joekytc/dsh-swarm

by maliang

自动团队:把一个目标拆成带依赖的任务图,按角色自动组队并行派发,每个子 Agent 独立模型,完成必须过门禁。

Workflow & AutomationManifest valid

★ 0

MIT

TypeScript

Nov 17, 2026

dsh plugin --profile web add dsh-auto-team