Back to directory

assay

A command-line tool that opens a web page in a real browser, works out a test plan purely mechanically from the controls it finds, drives them, and reports exactly which control broke and how.

assay

assay points at a web page, opens it in a real browser, measures every control it renders, works out a test plan from what it finds, drives all of it, and reports what broke. Because the plan comes from the page itself rather than from hand-written tests, a program written moments ago can be checked moments later with no baseline to maintain, and it uses no language model at all, so it gives the same answer on every run with no API key, tokens or rate limit.

It ships skills and plugins for fifteen coding-agent harnesses, including a Claude Code plugin that checks a page at the end of every turn that touched one. By default everything prints to standard output and nothing is written to disk; passing a report flag writes a self-contained HTML report with before-and-after screenshots, and a JSON flag prints the whole run for piping, with a non-zero exit code on any failure. It installs from PyPI and needs Python 3.10 or newer.

More in Testing

See all 69
  • Canary

    runcanary.ai

    An independent testing service, backed by Y Combinator, that runs an application and tries to break every change an AI coding agent makes, reporting what would have reached production.

  • MoEWatch

    github.com

    A diagnostic and audit library for Mixture-of-Experts models in HuggingFace Transformers that instruments router modules to catch expert collapse, routing entropy collapse and load imbalance during training.

  • Mockwave

    github.com

    An open-source, multi-protocol mock server for HTTP, GraphQL, SOAP, and gRPC, with weighted traffic splitting, dynamic JavaScript responses, chaos-testing fault injection, and an MCP server so an AI assistant can generate mock rules directly.