DeepSeek Harness plugin

metavibe_dsh

MetaVibe as a native DeepSeek Harness plugin: a read-only golden architecture map (hub) and best-practices catalog model tools for Vibe Coding.

Jump to install

Source facts

Repository
MetaVibeCoding/metavibe_dsh
Latest update
Aug 16, 2026
Category
Tools & Capabilities
GitHub stars
0
Format
plugin
Package path
metavibe-dsh
Catalog evidence
Upstream dsh.bundle evidence
Evidence path
metavibe-dsh/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/MetaVibeCoding/metavibe_dsh/tree/HEAD/metavibe-dsh
Plugin: metavibe_dsh
Author: MetaVibeCoding

Check the source files

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

File explorer3 files
README.mdSource · read only
README language

metavibe-dsh 🚀

Language: English | 中文

MetaVibe as a native DeepSeek Harness plugin — a read-only architecture advisor for Vibe Coding. The original Python CLI engine is retired; this package is the single implementation: TypeScript sources + Cordis plugin + defineTool model tools, built with the exact pipeline the official packages use (tsclib/types/, tsdownlib/index.js).

🧭 What it is (and what it is not)

MetaVibe advises — it never reaches into the project being worked on.

  • Architecture map (metavibe_hub_list): the built-in golden meta-architectures (layers / slots / guardrails) to pick a top-level design direction.
  • Best-practices catalog (metavibe_catalog_tree / metavibe_catalog_inspect): the knowledge matrix (data flows, data models, philosophies, meta-skills) with golden examples and agent instructions.
  • ❌ No workspace scanning, no file writes, no spec binding, no code generation, no guardrail enforcement inside the target project.

Because every tool is pure and read-only, the plugin needs no fs service, cannot stall the agent loop with workspace sweeps, and never interferes with the project it is advising.

🛠️ Tools (4 model tools)

ToolPurpose
metavibe_blueprintFlow-first advisory: classify the required information flows (write / read / event / integration / realtime / task), match the golden architecture(s), and compose a blueprint (layers / slots / guardrails) with alternatives and gap suggestions
metavibe_hub_listList the golden architecture map: name, source, version, description, flows, layers and slots per preset
metavibe_catalog_treeBrowse the knowledge matrix by category (data flows / data models / philosophies / meta-skills)
metavibe_catalog_inspectInspect one catalog entry in depth: summary, data-flow diagram, schemas, golden examples, agent instructions

Advisory method (flow-first): information-flow paths are the primary lens — metavibe_blueprint first recognizes the flows a system needs, then maps them onto common paradigms, then synthesizes an architecture blueprint. The hub's golden architectures each declare the flows they realize, so flow → architecture matching is grounded in the knowledge base.

🎯 Triggers & Usage Scenarios

How these tools get triggered inside a DeepSeek Harness session: the agent maps a user request to a concrete tool call. All tools are read-only and never touch the workspace.

When the user says…The agent calls
“What architecture should I use for a clean-arch web API?” / “帮我选个后端架构”metavibe_hub_list
“How do I structure CQRS / DTOs / an auth factory?”metavibe_catalog_treemetavibe_catalog_inspect
“Give me the golden patterns for payments / Next.js / FastAPI”metavibe_hub_list (pick the preset) → metavibe_catalog_inspect (best practice details)

Scenario — choose a top-level architecture direction

1. metavibe_hub_list — the agent inventories the golden architecture map. 2. The agent (and the user) pick the preset that fits the project, and the agent proposes the layer/one-way-dependency plan from the spec. 3. The agent drafts the project structure following the spec — the plugin only guides, it never writes files.

Scenario — look up a best practice while coding

1. metavibe_catalog_tree — overview of the knowledge matrix. 2. metavibe_catalog_inspect { "id": "data_flows/cqrs_flow" } — data-flow diagram, schemas, golden example code, agent instructions. 3. The agent applies the pattern in the code it is writing.

📁 Structure

metavibe-dsh/
├── package.json          # ESM package metadata (name: metavibe-dsh, main: lib/index.js)
├── tsconfig.json         # mirrors official base: es2024 / bundler / .ts imports → .js on emit
├── tsdown.config.ts      # same shape as official: entry lib/types/index.js → lib/index.js
├── cordis.yml.example    # mounting example (copy into an agent preset)
├── scripts/
│   ├── install.sh            # one-command installer (explicit dsh.bundle profile-layer plugin)
│   ├── assemble-dynamic.mjs  # assemble the session demo Package 1:1 from compiled output
│   └── gen-data.mjs          # regenerate src/data/hub.ts from skeletons/*.json
├── skeletons/            # golden meta-architecture sources (.json spec + .md design doc)
├── src/                  # TypeScript sources (every file < 300 lines)
│   ├── index.ts          # Cordis plugin entry (name/inject/Config/apply)
│   ├── engine.ts         # read-only engine (Hub map + Catalog matrix)
│   ├── specs.ts          # Spec types & parsing (lossless JSON: absent fields omitted)
│   ├── tools/            # grouped tool registration (hub / catalog / blueprint / helpers / index)
│   ├── data/             # embedded Hub / Catalog data (.ts)
│   └── types/dsh.d.ts    # ambient types for the cordis / dsh-tools runtime contract
├── tests/                # vitest suite (engine + tools, 13 cases)
├── examples/             # historical before/after effect-comparison projects (pre-0.3)
├── docs/                 # effect-comparison documentation (historical, pre-0.3)
└── lib/                  # build output (tsc → lib/types/, tsdown → lib/index.js)

All modules honor MetaVibe's own anti-entropy rules (single files < 300 lines). engine.ts + specs.ts are dependency-free pure logic (no I/O at all) and unit-testable standalone; tools/* only wires the contract.

🔨 Build

pnpm install       # devDeps: typescript / tsdown / @types/node / schemastery
pnpm test          # vitest
pnpm run typecheck # tsc --noEmit
pnpm run build     # tsc → lib/types/ + tsdown → lib/index.js

@deepseek-ai/dsh-tools / @deepseek-ai/cordis are peerDependencies supplied by the host deployment. The npm-registry versions of these packages are older than the runtime API, so they are NOT installed for type checking; src/types/dsh.d.ts declares the exact contract the plugin consumes.

📦 Install & Mount

Recommended — explicit profile-layer plugin. The package declares dsh.bundle (see package.json), so installing it with the one-command installer makes it a first-class plugin of the profile: dsh plugin add installs the package and automatically appends it to the profile's dsh.profile.bundles layer list. The plugin loads with the profile and the tools are available in every session — no manual patch editing, no agent preset to pick:

cd metavibe-dsh
bash scripts/install.sh                # web profile (default)
bash scripts/install.sh --profile tui  # a different profile

After installing, restart dsh web; metavibe_hub_list / metavibe_catalog_tree / metavibe_catalog_inspect appear in all sessions.

Per-session alternative (agent preset): if you only want the tools in one preset, copy the row from [cordis.yml.example](cordis.yml.example) into that preset's agent.cordis.yml instead. No config needed.

> 🚀 Publishing to the DSH plugin ecosystem (npm publish → dsh plugin add metavibe-dsh → mount) → see [PUBLISHING.md](PUBLISHING.md).

↔️ History

  • 0.4.0 — flow-first advisory: added metavibe_blueprint (classify information flows → match golden architectures → compose blueprint), a flows dimension on every hub architecture, and six data-flow primitives in the catalog.
  • 0.3.0 — scoped to a read-only architecture advisor: metavibe_check / metavibe_hub_use / metavibe_assemble / metavibe_inject / metavibe_extract_* were removed (they scanned, wrote to, or generated code in the target workspace). The plugin now consumes only the tools registry — no fs service, no config, no sandbox writes.
  • ≤ 0.2.x — the anti-entropy suite (guardrail check, spec binding, rule injection, slot assembly, extraction). See docs/effect-comparison.md for the historical before/after record.