DeepSeek Harness Plugin

SeveNOlogy7/oh-my-dsh

Category Workflow & Automation Added 2026-11-10 npm @jhcai11/oh-my-dsh

Agent orchestration for DeepSeek Harness with eleven selectable presets, planning, delegation, review gates and durable project state.

Install

# from npm (prebuilt)

dsh plugin --profile web add @jhcai11/oh-my-dsh

# from GitHub (first run asks for allowBuilds approval — follow the hint, retry)

dsh plugin --profile web add github:SeveNOlogy7/oh-my-dsh

Any plugin you install runs third-party code with your own permissions — it can read your files, use your credentials, and reach the network, and tool approvals don’t sandbox it. GitHub-sourced plugins also run build scripts at install time — pnpm blocks those until you allow them, so an install can stop with ERR_PNPM_GIT_DEP_PREPARE_NOT_ALLOWED or ERR_PNPM_IGNORED_BUILDS; dsh prints the exact key to add under allowBuilds in your profile’s pnpm-workspace.yaml, and the install works on the next run. Allowing a build is a trust decision: only install sources you trust, and pin a commit (github:owner/repo#sha).

README

English | 简体中文

Agent orchestration for DeepSeek Harness, aligned with oh-my-openagent v4.19.4 (commit b072d27).

Oh-My-DSH installs eleven selectable Agent presets and a durable .omd/ orchestration state: planning, interviews, delegated workers, background tasks, review gates, evidence, and handoff. It is a clean-room capability port: no upstream code or prompt text is copied. See NOTICE and upstream/omo-sources.lock.json.

Install from npm

npm install @jhcai11/oh-my-dsh

or add the bundle straight into a DSH profile:

dsh plugin --profile <profile> add @jhcai11/oh-my-dsh

The bare name oh-my-dsh belongs to an unrelated package, which is why this one is scoped. docs/releasing.md records the release path: the scoped rename, npm publish --access public, publishing later versions from CI with provenance, and the market submission.

Install

From this checkout, install the bundle into the current DSH profile:

plugin_manager install_bundle  target: E:\Oh-My-DSH

Or from a terminal:

dsh plugin --profile <profile> add E:\Oh-My-DSH

The install adds only eleven agent-preset declarations. It does not add global tools, commands, hooks, or a default preset. Standard sessions are unchanged. If the app was already running, restart it once before selecting an OMD preset so the current runtime module generation is loaded.

Language

The language switch for OMD preset labels and messages is an OMD configuration concern, not an install-time flag. The complete description lives in docs/configuration.md.

Use

  1. Start a new session and select one of the OMD presets, for example Sisyphus · OMD. Presets are chosen before the first turn.
  2. Run /omd-init to create .omd/.
  3. Run the omd-doctor skill to verify the composition.
  4. Use /omd-ulw <goal>, /omd-plan <goal>, /omd-start-work, or just ask.
Preset Role
Sisyphus · OMD orchestrator
Hephaestus · OMD autonomous deep worker
Prometheus · OMD interview planner
Atlas · OMD plan conductor (read-only)
Oracle · OMD read-only architecture consultant
Librarian · OMD read-only docs/OSS researcher
Explore · OMD read-only codebase scanner
Multimodal Looker · OMD read-only visual analyst
Metis · OMD plan gap analyst
Momus · OMD adversarial reviewer
Sisyphus-Junior · OMD delegated task executor

Commands

/omd-init, /omd-status, /omd-config, /omd-ulw, /omd-plan, /omd-hyperplan, /omd-team, /omd-start-work, /omd-stop-continuation, /omd-handoff, /omd-init-deep, /omd-refactor, /omd-remove-ai-slops.

/goal is provided by DeepSeek Harness itself.

The doctor is a skill rather than a command: every OMD role carries omd-doctor, and it drives the omd_doctor tool so the diagnosis and the repair happen in one turn. See docs/commands.md and docs/tools.md.

Settings → Oh-My-DSH is the bundle's settings page: it reads and writes the session workspace's .omd/config.jsonc and reports the effective language and routing tiers with their source. See docs/configuration.md.

Tools

omd_delegate, omd_background_output, omd_background_cancel, omd_background_list, omd_session_list, omd_session_read, omd_session_search, omd_task_* (opt-in), omd_notepad, omd_lsp, omd_diagnostics, omd_look_at, omd_mode, omd_status, omd_doctor.

Zero intrusion

  • OMD tools/commands/hooks/skills exist only inside OMD presets.
  • Installing the bundle changes no default preset and writes nothing to .omd/ until an OMD session uses a command or tool.
  • Keyword activation (ulw, team, hyperplan) is on inside OMD presets and can be disabled in .omd/config.jsonc.

Verify

npm test
npm run check

Then in a new OMD session run the omd-doctor skill: it calls the omd_doctor tool and reports PASS/WARN/FAIL with a fix for each failure. See docs/install-verify.md and the recorded docs/verification.md for the mount-probe evidence and the remaining manual gate.

Scope for v0.1.0

The orchestration loop is complete: eleven roles, category and named-agent delegation, background queue and results, plans, boulder slots, notepads, review verdicts, sessions, tasks, rules, model routing, and the ultrawork, plan, start-work, hyperplan, team, review, handoff, init-deep, refactor, and AI-slop protocols.

Host-bound OmO 4.19.4 features are mapped or deferred with reasons in docs/parity-omo-4.19.4.md: dashboard/Client UI, Team Mode's team_* tool family, hashline edit, interactive tmux, skill-embedded MCP, CodeGraph, Claude Code hooks, and dynamic model fallback chains.

License

MIT for this repository. The upstream project is SUL-1.0; see NOTICE and upstream/.

Content from the project README on GitHub ↗

Links

More in this category

View the whole category →

Community comments

Comments are public GitHub Discussions. Loading them connects to GitHub and Giscus; a GitHub account is required to post.