Core Concepts
The pieces Recus uses to turn one URL into a reviewable first pass of QA coverage.
Recus works with your tester
The scan model
Starting URL
↓
Page context + product brief
↓
Elements → first-pass test cases
↓
Skill decisions → browser actions + evidence
↓
Discoveries → reveal paths → selected same-host pages
↓
Pass / fail / warning → human review → reportBuilding blocks
Scan
One run against one starting URL. A scan contains the page study, generated test cases, browser execution, discoveries, crawl results, evidence, and final report.
Page context and product brief
Before interacting, Recus combines a full-page visual read with the accessibility tree. It writes a short brief describing what the page appears to do, who it serves, and what successful behavior should look like.
Element
An interactable control found through vision and accessibility data: inputs, buttons, links, tabs, menus, and controls revealed below the fold. Each element keeps its accessible name and locator context.
Test case
A generated check against one element or behavior. Recus drafts happy paths, empty submits, invalid formats, boundaries, security probes, keyboard checks, and UX assertions for your team to review and extend.
Discovery and reveal path
Interactions can expose new UI such as dialogs, menus, and extra form steps. Recus records the exact actions that revealed each element, then replays that path before testing it.
Crawl
Bounded exploration of selected routes on the starting hostname. Recus checks scope, repetition, session needs, and page importance before deciding which destinations receive their own study and tests. Discovered does not mean tested.
Verdict
A pass, fail, or warning produced from the visual result, action log, and browser state. The verdict is a recommendation backed by evidence; your reviewer makes the final call.
Report
The shareable artifact from a scan: findings, before-and-after screenshots, video, Playwright traces, action logs, discovered routes, explanations, and suggested fixes.
Human in the loop
Some tests need authorized input or access the agent does not have. Recus can ask for human input and use it for the affected test. Missing input can leave a check skipped or unverified; not every blocker can be resolved automatically.
Skill and skill decision
A skill provides a capability beyond ordinary browser actions. Email is the current enabled skill. Recus selects skills for individual tests and distinguishes full, partial, and unavailable capability coverage. Full skill coverage is not a passing verdict.
How to read a verdict
Pass
Observed behavior matched the generated expectation.
Fail
Evidence suggests a functional, validation, or security defect.
Warning
Uncertainty, incomplete verification, or usability concerns need review.
Treat the report as a first pass