Autonomous Browser QA

AI-led QA loop

Give Impact Gate a diff, a feature name, or a staging URL, and it can take the lead on browser-side QA: scope the run, explore the app, hunt bugs, generate follow-up tests, heal failures, and return an evidence-backed verdict. This is the product story for replacing manual smoke passes and ad hoc exploratory testing with one autonomous workflow.

What it replaces

Manual QA work turns into one guided autonomous loop

hand-built smoke passes
click-by-click exploratory testing
ad hoc bug hunts
first-draft test authoring
retrying brittle specs until they pass

The loop

One prompt in, browser evidence out

diff or prompt -> scope with route families and traceability -> explore in a real browser -> find issues and capture evidence -> generate or heal tests when needed -> verify and report

Start here Modes From findings to tests Trust what it finds What you get

Start Here

Natural language

Tell the agent what to validate and let it choose the path

/qa test this app at http://localhost:3000 /qa hunt "checkout flow" on https://staging.example.com /qa run a release regression on https://preview.example.com /qa verify the healed tests only

The /qa skill is the natural-language front door. It maps the request to the right browser-side workflow so the AI can lead the charge instead of making you assemble flags by hand.

Direct CLI

Use the engine directly when you want the full browser loop now

npx impact-gate-qa pr --since origin/main --base-url http://localhost:3000 npx impact-gate-qa hunt "checkout flow" --base-url http://localhost:3000 npx impact-gate-qa release --base-url http://localhost:3000 --time 30 npx impact-gate-qa fix --base-url http://localhost:3000

If you already know the mode you want, the CLI gives you the same autonomous browser engine without the wrapper.

Modes

PR mode

Diff-scoped validation for changed features

tests the features that changed in the branch or PR
uses the diff to stay focused on the real blast radius
returns findings, evidence, and a verdict for merge decisions

Hunt mode

Deep exploratory testing for bugs and edge cases

attacks a named area like checkout, billing, or settings
lets the AI roam beyond scripted happy paths
is built for bug hunts, edge cases, and surprise regressions

Release mode

Broader autonomous regression before you ship

systematically covers critical flows for a release candidate
uses the browser loop to gather evidence while it explores
produces a go / no-go / conditional style outcome

Fix mode

Verification for healed specs and repaired paths

re-runs the browser path after repairs or spec updates
confirms the fix before it is trusted again
keeps the loop closed instead of leaving a half-finished repair behind

From Findings To Tests

Generation

When the browser finds a gap, the same stack can draft the fix

npx impact-gate generate --path . --max-attempts 3 npx impact-gate heal --path . --traceability-report ./playwright-report.json npx impact-gate finalize-generated-tests --path . --create-pr --pr-title "Add E2E tests"

Browser exploration is not the end of the story. Impact Gate can turn the discovery into runnable specs, run them, and heal them until the code and the evidence agree.

Full autonomy

Let the crew carry the flow from design to generated and healed specs

npx impact-gate crew --workflow full-qa --path . --tests-root ./e2e-tests --since origin/main npx impact-gate analyze --path . --generate --heal

Use the full-qa workflow when you want the system to keep moving through design, generation, execution, and healing with minimal manual effort.

Trust What It Finds

Scope control

Deterministic context keeps the browser loop focused

route families map changed code to product areas
traceability data tightens the test-to-file link
impact and plan decide where the AI should spend time
the browser loop starts with evidence, not guesswork

Trust controls

Generation and repair stay guarded

generation uses discovered API surface instead of invented helpers
suspicious specs are quarantined instead of trusted automatically
compiled and smoke-verified output stays in the trusted path
browser evidence and reports stay visible for review

What You Get

Evidence

Findings that are easy to act on

screenshots and console errors
step-by-step browser findings
health scores and verdicts
release-readiness context

Artifacts

Outputs you can keep using in CI

.e2e-ai-agents/qa-summary.md
.e2e-ai-agents/qa-report.json
.e2e-ai-agents/plan.json
generated and healed Playwright specs

Bottom line

This is the browser QA loop you reach for when you want AI to do the heavy lifting, reduce manual effort, and keep moving until it has both evidence and runnable code.

QA Skill Crew Workflows AI Guardrails Route Families CLI Reference