Writing tools that modify a page and build it does not prove that the page has loaded, interactions work, the browser is error-free, or the result matches expectations. dsh-live-loop provides a coherent runtime loop for an installed DeepSeek Harness Agent.

What Is This

This is a frontend runtime verification plugin for DeepSeek Harness. It enables an Agent to detect, run, preview, interact, visually compare, and verify a web application while generating persistent evidence.

  • Maintainer: POWERRRRRRRR
  • License: MIT
  • Positioning: Handles the development server lifecycle, isolated browser state, structured observation, interactions, verification, and persistent evidence. It does not edit application code; DSH’s normal coding tools remain responsible for code modifications.

Installation and Enablement

Clone the repository from GitHub and install it into the web profile.

git clone https://github.com/POWERRRRRRRR/dsh-live-loop.git
cd dsh-live-loop
npm ci --legacy-peer-deps
npm run build
npm pack --ignore-scripts
dsh plugin --profile web add ./dsh-live-loop-1.0.0.tgz

After installation, restart the web profile to ensure the plugin takes effect:

dsh --profile web web

Core Features

  1. Full-cycle loop: detection, running, preview, interaction, visual comparison, and verification, with persistent evidence throughout.
  2. Structured detection: supports Vite, React, Vue, Next.js, generic scripts, and monorepos, with explicit ambiguity resolution to avoid silently choosing the wrong run target.
  3. Process ownership: provides argv-only execution through the public DSH subprocess ownership, including bounded logging and cleanup mechanisms.
  4. Isolated environment: each DSH topic has its own independent BrowserContext.
  5. Bounded observation: bounded observation windows for HMR, WebSocket, SSE, and polling.
  6. Evidence management: generates DSH Attachment screenshots with Reference, Current, and Diff evidence.
  7. Strict verdicts: uses a strict four-state verdict mechanism and rejects converting missing observations into passes.
  8. Native integration: provides native DSH Web Live Preview, Verification Card, and Settings panels.
  9. Diagnostics: provides diagnostic information when verification fails, explicitly telling the Agent what needs to be fixed.

Typical Usage

An Agent can complete the verification workflow through tool calls. For example, fix a form and verify specific behavior:

Fix the form, start or reuse the detected app, fill the Name textbox with Ada,
press Enter, assert that “Hello, Ada!” appears, and do not finish until
live_loop_verify returns VERIFIED with screenshot evidence.

A typical failure handling loop is as follows:

first verify: FAILED
→ inspect Console / Network / DOM / assertion / visual diff
→ edit the application
→ reload or let HMR apply
→ verify again: VERIFIED
→ deliver the report and evidence

Applicable Scenarios and Notes

  • Environment requirements: requires DeepSeek Harness version 0.1.0-rc.7, Node.js version ^22.19.0 or >=24.
  • Code editing: the plugin itself does not edit application code; DSH’s coding tools are responsible.
  • Evidence weight: visual similarity is supporting evidence and never overrides the results of page loading, diagnostics, interactions, or assertions.

Summary

dsh-live-loop provides reusable runtime verification capabilities for frontend agents by encapsulating browser operations and verification logic as DSH tools. It ensures the rigor of the verification process and the certainty of results, and is a necessary component for building reliable frontend automation workflows.

  • GitHub: https://github.com/POWERRRRRRRR/dsh-live-loop
  • Catalog: https://www.skillhub.cn/plugins/POWERRRRRRRR/dsh-live-loop