DeepSeek Harness plugin

dh-workspace

Out-of-tree dsh plugin bundle: philosophy skills plus workspace, delegation, subagent-preset, and worktree plugins (naming contract in DESIGN.md)

Jump to install

Source facts

Repository
atesahmet0/dh-workspace
Latest update
Aug 17, 2026
Category
Workflow & Automation
GitHub stars
1
Format
plugin
Catalog evidence
Upstream dsh.bundle evidence
Evidence path
package.json#dsh.bundle
Checked against
0.1.0-rc.8
Upstream check date
2026-08-20

This evidence comes from the upstream catalog. This site has not installed, run, or security-reviewed the plugin.

Install

Start with a prompt that asks an agent to review the GitHub repository and source. Switch to the command if you want to install it yourself.

Copy this prompt into DSH, Codex, or another agent and ask it to review the GitHub repository and source first.

Do not install or run any commands yet. Read this plugin's GitHub repository, README, and relevant source code. Then answer the questions below clearly and directly so I can decide whether it fits my needs:

1. What is this plugin, and what problem does it solve?
2. Who is it for, and what are its typical use cases?
3. How is it used after installation? Include one minimal example.
4. What known limitations or privacy, security, compatibility, or maintenance risks does it have?
5. Give a clear recommendation: recommend, conditionally recommend, or do not recommend, with reasons.

Distinguish statements documented by the repository, inferences from source code, and unknowns. If evidence is insufficient, say so explicitly. Do not guess or simply repeat the README.

GitHub: https://github.com/atesahmet0/dh-workspace
Plugin: dh-workspace
Author: atesahmet0

Check the source files

Read the README and other files from this plugin directory before installing.

File explorer3 files
README.mdSource · read only

dh-multiagents

Multi-agent orchestration for the DeepSeek Harness (dsh). A team of role-bound agents — plan, build, explore, researcher, coder, scribe, reviewer — with tools enforced by a runtime capability matrix.

Why dh-multiagents?

  • Plan-first — read-only orchestrators delegate, never write code themselves.
  • Enforced, not suggested — each preset's tool allow-list is applied via tools.restrict, so a plan agent physically can't run bash.
  • Auditable delegation — every delegate call persists an id (e.g. daring-pearl-elk) with a result file on disk.
  • Zero-config install — one command, presets + skills wired automatically.

Installation

Requires Node ≥ 22.19 and a dsh install (@deepseek-ai/dsh@0.1.0-rc.6).

dsh plugin --profile <name> init
dsh plugin --profile <name> add @dh-multiagents/bundle

> Presets are mirrored automatically at every boot (the plugin copies them > into $DSH_HOME/.agent-presets/), so there is no install-time step.

> Fresh dsh CLI (npm 11): npm 11 blocks dependency install scripts by > default — the harness's own native deps (node-pty, dsh-subprocess-local, > koffi, protobufjs) are gated too. Run npm install-scripts approve for > them before headless mode will boot.

Quick Start

> [!IMPORTANT] > Select Plan or Build at the start of your session: > In the Web or TUI interface, select either the plan or build preset in the preset picker at the beginning of your session: > - plan: Use when starting a task to research the codebase and write an implementation plan (delegates to explore and researcher). > - build: Use to execute an approved plan (delegates to coder, scribe, and reviewer with worktrees). > > Only plan and build act as orchestrators. The subagent presets (explore, researcher, coder, scribe, reviewer) are spawned automatically via delegate and should not be selected as top-level sessions.

# headless mode needs the headless bundle (add once)
dsh plugin --profile <name> add @deepseek-ai/dsh-headless

DEEPSEEK_API_KEY=... dsh --profile <name> headless \
  "Use plan_save to save a plan titled '# P' for project 'demo'. Then delegate '1+1?' to the researcher and report its answer."

For full orchestration (delegate/build), use the tui or web profile — headless sessions are preset-less by design, so delegate is rejected there (only plan/build presets may delegate).

What you get

PresetRoleNotable tools
planRead-only orchestratorplan_save/read, delegate → explore/researcher
buildRead-only orchestratordelegate → coder/scribe/reviewer, worktrees
exploreCodebase explorerread, glob, grep
researcherWeb researcherweb_search, read
coderImplementerbash, edit, write, worktrees
scribeDocs writeredit, write, read
reviewerRead-only reviewerread, glob, grep

Plus 5 philosophy skills (code-philosophy, code-review, plan-review, plan-protocol, frontend-philosophy) loaded into every session.

Runtime behavior

  • Delegations persist as $DSH_HOME/workspace/<project>/delegations/<id>.md + <id>.result.txt — auditable after the fact.
  • Only plan/build may delegate; plan → explore/researcher, build → +coder/scribe/reviewer. No delegation from inside a subagent (anti-recursion).
  • Children inherit the parent's model selection.

Development

pnpm install && pnpm -r build
  • Naming contract: identifiers are locked in DESIGN.md — change it there first, never in code alone.
  • Presets/skills: add presets/<name>/ + a CAPABILITY_MATRIX entry (dh-common), or skills/<name>/SKILL.md (dh-philosophy).
  • Publishing: bump versions, then pnpm pack and publish the tarballs — plain npm publish leaves workspace:* unresolved (broke 0.1.0).

License

MIT