dsh-svw-waveform

dsh-svw-waveform

Native SVW waveform rendering for DeepSeek Harness.

The plugin registers svw_wave_render. The model supplies an exact VCD or FST path, native tick range, and one to twelve hierarchical signal names. In the DeepSeek Harness Web UI the result becomes an interactive waveform viewer with zoom, pan, markers, radix selection, edge navigation, and delta-cycle markers. Surfaces without the client plugin fall back to the complete ANSI terminal waveform card.

Install

Install the public npm bundle into the Web profile:

dsh plugin --profile web add dsh-svw-waveform
dsh --profile web --dump-config
dsh --profile web

The npm bundle selects a package-private SVW executable for Linux x64 or macOS arm64. It does not compile on the consumer machine and has no install, postinstall, or prepare script. Set SVW_BIN=/absolute/path/to/svw to use an independently installed binary instead. On macOS, an existing Homebrew SVW installation is also accepted as a fallback.

DeepSeek Harness is still a developer preview. This package currently supports the @deepseek-ai/dsh 0.1.x release-candidate line and tests every change against a pinned version before publication.

Use

First obtain exact signal names:

svw agent design.vcd signals reset 20

Then ask the agent to display a bounded window, or call the registered tool with arguments equivalent to:

{
  "waveform": "design.vcd",
  "start": 100,
  "end": 160,
  "hier": ["top.clk", "top.reset_n"],
  "width": 120,
  "height": 24
}

The bundled svw-waveform skill teaches the agent the complete discovery, value/change, render, comparison, semantic-design, UVM, and coverage workflows.

Artifact provenance

SVW GitHub Releases are the only binary source. This repository never rebuilds SVW. Its release workflow:

1. downloads the exact release-X.Y.Z Linux and macOS archives and SHA256 sidecars;
2. rejects unsafe archive paths and extracts only declared regular files;
3. smoke-tests svw --version, waveform metadata, and a real render;
4. audits dynamic libraries on clean GitHub-hosted runners;
5. packages the byte-identical bin/svw files into platform-specific npm packages before publishing the entry bundle.

The platform packages are:

• dsh-svw-waveform-linux-x64
• dsh-svw-waveform-darwin-arm64

Each package includes svw-provenance.json with the source Release asset and both archive and binary SHA256 values. Replacing an asset under the same SVW tag creates a new immutable plugin patch version rather than overwriting an npm version.

The host and browser integration are synchronized from the versioned share/svw/agents/dsh-extension files inside the same verified SVW archive. svw-source.json records their original hashes and the small downstream packaging transforms.

Publishing setup

The publish.yml workflow accepts either repository_dispatch event svw-release-published or a manual release-X.Y.Z input. It supports a temporary NPM_TOKEN for bootstrapping the three package names. After that, configure npm Trusted Publishing for each package with:

• GitHub organization: svcomplex-dev
• repository: dsh-svw-waveform
• workflow filename: publish.yml
• allowed action: npm publish

Once all three trusted publishers work, remove NPM_TOKEN. The workflow uses GitHub OIDC and publishes npm provenance automatically.

Development

npm run verify
sh scripts/setup-git-hooks.sh

All commits and annotated package tags must use svcomplex-dev <code@svcomplex.ai>.