Browser verification
for AI-generated
web apps
Agents write UI code. Verfix verifies it works.
Structured JSON for agents. Execution timeline for humans.
npx verfix init The problem
Stop manually clicking through flows
AI agents generate entire application flows in seconds. Verifying those flows still requires a human.
How it works
Three steps from zero to verified
One command to set up. One command to run. Agents get structured output. Humans get full observability.
STEP 01
verfix init
Configure in 60 seconds
The wizard starts the Docker runtime, detects your app's URL, scaffolds a
verfix.config.json flow library, and generates
AGENTS.md so coding agents know how to use Verfix.
STEP 02
verfix run --flow <id>
Agent-driven verify loop
Every UI change triggers a flow run. Playwright executes it in an isolated browser
context. The agent receives passed,
failures[], and fix_hint.
It fixes and retries autonomously.
STEP 03
open timeline_url
Human-readable timeline
If the agent can't self-fix, timeline_url surfaces
the full execution — every action, navigation, console error, network request, and
failure screenshot.
Execution modes
Right tool for every stage
Each flow can override the global mode — stable flows use strict, evolving ones use assisted.
Strict
Fully deterministic. CSS selectors and data-testid
only. No AI fallback. If a selector breaks, it fails explicitly — no ambiguity.
Best for: CI/CD, regression testing
Assisted
Deterministic first, AI fallback if selectors break. Tries selector → semantic discovery → AI resolver before failing. Self-healing without compromising correctness.
Best for: Active development, new flows
Exploratory
Natural language task. The AI navigates, reasons, and verifies on its own. Optimized for discovery. Clearly marked in output — not CI-grade.
Best for: Ad-hoc QA, unknown flows
Quickstart
Up in four commands
Requirements: Docker Desktop (running), Node.js 18+
Install
npm install -g verfix Init your project
npx verfix init The wizard detects your app URL, creates verfix.config.json, and writes AGENTS.md.
Run a flow
verfix run --flow login --output json Returns structured JSON. Exit code 0 on pass, 1 on failure.
Open the timeline
# Copy timeline_url from output → open in browser
Full execution replay: every action, console log, network request, and screenshot.
Default: http://localhost:3610/?executionId=exec_...