jev-browser-qa
Browser QA where Playwright drives and records while Jev judges assertions, matches controls to intent and picks goal-directed clicks.
Overview
jev-browser-qa splits end-to-end testing between Playwright and Jev. Playwright handles navigation, typing, waiting, video, traces and screenshots; Jev handles judgment: whether a natural-language claim holds on the page (expect), which visible control matches an intent (click or pick, giving self-healing locators), and what the next click toward a goal is (goal and runGoal). Text entry is never delegated to the model, and Jev sees text such as the URL, title, visible text and accessibility tree rather than pixels.
It offers a JSON-flow CLI for agents that runs a flow and returns one JSON result with pass or fail per step, the probability behind each judgment, a video and a final screenshot, with exit codes and secrets referenced as placeholders; primitives for use in Playwright tests; a run dashboard; and an agent skill. Jev is accessed through the Vercel AI Gateway via the AI SDK's experimental evaluate, so one gateway key is the whole configuration. The README cites answers in under 400 ms with no output tokens. It has CI and an MIT licence.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Yes / no | Does this claim hold on the page? | yes · no |
| Choice | Which visible control matches this intent? | — |
| Choice | What is the next click toward this goal? | — |
How it works
- Write a JSON flow or Playwright test with natural-language steps
- Playwright navigates and records
- Jev evaluates assertions, locates controls or chooses next clicks
- Return per-step pass/fail with probabilities, video and screenshot
Strengths
- Natural-language assertions with probabilities
- Self-healing locators by intent
- Single gateway key configuration
Caveats
- Model judgments can be wrong; thresholds matter
- Depends on Vercel AI Gateway access
At a glance
- Type
- Agent skill
- Status
- beta
- Language
- TypeScript
- Runtime
- Playwright, Node CLI, dashboard
- Has code
- yes
- Has tests
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
Browser QA where Playwright drives and films, and TypeSafe Jev judges. JSON-flow CLI for agents, run dashboard, agent skill.
logicrw/awesome-jev-projects
Frequently asked questions
What does Jev judge?
Whether assertions hold, which control matches an intent, and the next click toward a goal.
Does Jev type text?
No, text entry is always done by Playwright.
How is Jev accessed?
Through the Vercel AI Gateway with one API key.
Related projects
CodexQA browser skill that indexes in-page controls and has Jev pick the operation and target each step, with no screenshots needed.
Agent skillbetaDrop-in proxy for Playwright MCP that adds Jev page triage, prompt-injection masking, snapshot pruning and risky-action gating.
MCP serverbetaTypeScriptExperimental browser-only MCP server where Jev picks a code-owned action ID and a human approves before execution.
MCP serverprototypeTypeScriptPlaywright TypeScript project using Jev for semantic assertions, element picking, goal-driven steps and test failure classification.
ProjectprototypeTypeScriptPlain-English Playwright end-to-end tests where Jev resolves each step's target element and judges whether verify claims hold.
CLIprototypeTypeScriptEvidence-driven frontend QA where Jev picks browser actions while code checks DOM, HTTP and database contracts.
AppbetaPython