AI Agent Hub
Back to plugins
🖥️

dsh-visual-acceptance

Client Updated 2026.08.26

Run the following command in DeepSeek Harness:

dsh plugin install PolinniZhong/dsh-visual-acceptance

Paste the following prompt into your AI chat to install this plugin:

Run dsh plugin install PolinniZhong/dsh-visual-acceptance in your DSH Web Profile to install; the full source repository is at https://github.com/PolinniZhong/dsh-visual-acceptance

About this plugin

After an AI agent generates or rewrites a local web page, the gap between "it looks fine" and "it is actually fine" is often one real browser check away. DSH Visual Acceptance fills that gap without making aesthetic calls or editing code for you. It puts the declared-scope browser facts — Ready state, runtime failures, responsive quality signals, horizontal overflow, and validated screenshots — on your desk so you can accept or reject with evidence rather than gut feel.

Coverage plans range from a two-checkpoint Quick pass to an eight-checkpoint Custom matrix that auto-discovers the page CSS breakpoints and produces a cross-viewport overview. After a change you re-test against the same immutable Matrix; Baseline, Current, and Diff sit side by side, candidate relations (new-candidate, still-detected, possibly-resolved, regressed) are tagged automatically, and you confirm or reject each one by hand. Every step in the loop is human-gated: findings appear as Candidates, Issues are created only when you click, and change packages are copied only after you approve a fix — nothing crosses the human decision boundary on its own.

If you build and iterate local web products with DSH or Codex and need a traceable acceptance step between "the agent says done" and "I ship it," this plugin is that step. It runs on Node.js ≥ 22.19 inside a DSH Web session, keeps every artifact under the workspace .dsh-visual-acceptance/ directory, never touches public URLs, never uploads credentials, and never modifies project files or sends packages to an Agent.

Use Cases

  • Run a cross-viewport acceptance check after AI edits a local page
  • Retest under the original matrix and compare baseline, current, and diff screenshots
  • Log traceable browser failures and decide whether to create a tracked issue

Best For

  • Developers iterating local web products with DSH or Codex
  • Engineers who need traceable acceptance records after AI-generated frontends
  • Teams adding a human-gated checkpoint between agent delivery and launch