Preface

After generating local web pages with DeepSeek Harness (DSH) or Codex, acceptance work usually relies on manual screenshots or external tools, making it difficult to reproduce the state before changes. DSH Visual Acceptance is a local acceptance and re-verification workbench for AI-generated web products and UI prototypes.

Product Positioning

This is an “acceptance and re-verification workbench” plugin maintained by PolinniZhong (MIT License). It focuses on verifying in a local environment whether a page truly opens within the declared scope, recording traceable browser anomaly signals, and supporting re-verification under the original conditions.

Core Features

  1. Scope Validation: Checks whether the page truly opens within the declared scope.
  2. Anomaly Recording: Records traceable browser anomaly signals.
  3. Conditional Re-verification: Supports re-verification under the original conditions without external dependencies.
  4. Localization: Does not access public URLs, and does not perform reference image comparison, OCR, or visual model inference.
  5. Human-in-the-Loop: Does not automatically create formal Issues, modify project files, or send modification packages to an Agent; it only provides candidate Findings and entry points for manual decision-making.

Installation and Activation

In a DSH Web Profile, run the following commands to install the Beta version:

dsh plugin --profile web add github:PolinniZhong/dsh-visual-acceptance#v0.1.2-beta.1
dsh --profile web --dump-config
dsh web

dsh plugin add installs the Bundle and adds it to the current Profile; the dsh-visual-acceptance layer should appear in --dump-config. After restarting, open the “Visual Acceptance” Tab at the top of any non-empty conversation.

Uninstall command:

dsh plugin --profile web remove dsh-visual-acceptance

Typical Usage

  1. First start the local page, for example http://127.0.0.1:4173.
  2. Open “Visual Acceptance” in a non-empty DSH conversation.
  3. In “Page URL or File”, enter the local URL or an HTML file in the current workspace, for example examples/acceptance-demo/index.html.
  4. Select Quick, Standard, Page Breakpoints, or a custom plan, review the generated Matrix in “Confirm Acceptance Instructions”, and then click “Check N”.
  5. Review the Ready status, deterministic anomaly signals, and screenshots on the right. Note: Ready is satisfied and 0 anomaly signals are not equivalent to visual acceptance; they only indicate that corresponding anomalies were not detected.
  6. For Candidate items that need tracking, click “Create Issue”, add impact details, and save; you can also add a manual visual Issue from a screenshot.
  7. In “Issues and Decisions”, append manual Decisions; only after “Approve Modification” can you click “Copy Modification Package”.
  8. After completing modifications, click “Re-verify Under Original Conditions”, and manually confirm or reject based on the candidate relationships.

Notes

  • The current version is Beta.
  • Empty DSH conversations do not mount conversation.view; you need to send a message first before the Tab appears.
  • It does not make final aesthetic judgments for the user, and does not perform visual model inference or single quality scoring.
  • It does not automatically create formal Issues, modify project files, send modification packages to an Agent, or automatically write resolved.
  • It does not declare business status on behalf of the user.