dsh-verify

When an AI agent claims to have built a web app and declares it 'done,' can we truly trust its self-assessment? Traditionally, agents may rely on assumptions or static checks to validate outputs, but they often fail to simulate real user experiences in a browser—like clicking buttons, verifying style changes, or detecting console errors. dsh-verify addresses this credibility gap by acting as an independent quality gate that uses real browsers (such as Chromium, Firefox, or WebKit) to perform human-style checks, ensuring apps actually work as intended.

The core capability of this plugin lies in its deterministic verification: you simply write a JSON spec defining key interactions and expected outcomes, and dsh-verify automatically executes these steps in a browser, generating an HTML report with screenshots and diff images for clear pass/fail results. It supports multiple use cases, including serving as an MCP server for AI agents to self-verify during development, a CLI for manual or CI/CD pipeline checks, and integration into GitHub Actions for automated validation. Whether you're building frontend features or complex interactions, dsh-verify catches issues that agents might miss, such as missing CSS rules or JavaScript bugs.

dsh-verify is ideal for AI agent developers, teams using tools like DeepSeek Harness, and any engineer building web apps in CI/CD pipelines. If you're tired of agents 'grading their own homework' and want a reliable way to prove app quality, this plugin provides objective evidence. It avoids LLM judgment, instead letting the browser be the ultimate judge, helping you catch real-world defects before release to boost delivery credibility and user satisfaction.