Kane CLI now takes a requirement all the way to a ship-or-hold verdict, designing the tests, running them in a real browser, and emitting an open .evidence proof pack that teammates, AI agents, and auditors can all verify, with no server or dashboard required

TestMu AI (formerly LambdaTest), the world’s first Agentic AI-powered Quality Engineering platform, introduced a source-to-verdict loop in Kane CLI, its natural-language testing tool. Building on Kane CLI’s evolution from a browser automation tool, the loop carries a product requirement to a ship decision, writing the tests, running them in the local browser, collecting the proof, measuring coverage from what actually happened, and returning a verdict.

As AI agents write, run, and fix tests, a green checkmark is no longer enough: a step that checks nothing passes, a test written against a spec that changed weeks ago passes, and an agent that says “done” without the click ever landing passes. Empty green and earned green are the same color on every dashboard, the number climbs while the risk does not move. Kane CLI closes that gap: it carries a requirement to a verdict, and proves every step of it.

One loop. Source to verdict.

You no longer hand Kane CLI a test script, you hand it your requirements. Nine stages form one traceable line:

Source → Business use case → Scenario → Acceptance criteria → test.md → Execution → Evidence → Coverage → Verdict

The first half is the AI doing the test-engineering; the second half is the proof. The number at the end is read off the machine, never typed, never assumed.

• Ingest anything. Point Kane CLI at a PRD, Jira ticket, Confluence spec, Figma frame, or demo video and it pulls them into context — no other tool takes a Figma node or an mp4 as a requirement source.
• AI designs the tests, and asks when it should. It extracts business use cases, breaks each into scenarios, and pins every scenario to acceptance criteria. When two sources disagree, it stops and asks which is current, rather than guessing.
• Tests you can read, carrying their own coverage map. The output is test.md: plain Markdown a human can edit, where every step declares which criterion it proves. Traceability is built in, not a spreadsheet beside the tests.
• Deterministic execution in a real browser. Authoring is loose and intent-based; execution replays a recorded run with no LLM in the loop, so it behaves the same every time. Autoheal kicks in only when the product has genuinely drifted, and flows can run past 50 steps.
• One .evidence pack per run. Not a screenshot and a log, but the whole run: the agent’s trajectory, the network as a HAR, the DOM where things diverged, the console, the video, and a root-cause writeup of any failure. It opens as a page with levels — L0 minimal, L1 coverage and quality signals, L3 signing, attestation, and per-file hashes.
• Three scores behind every tick. Evidence-backed (was the pass real), legitimacy (did the agent really do it — a no-op click reads legit:false), and determinism (does it reproduce).
• A coverage number allowed to fall. Coverage is computed from the pack, not asserted by the tool that ran it. Strict mode demotes tests that passed against a since-changed spec — a percentage that can fall is the only kind you can trust when it rises.
• Then someone signs. Coverage, confidence, uncovered criteria, and stale sources roll up into one verdict, ship or hold, posted as a required check on the pull request, with the evidence link attached. The machine brings the proof; a person makes the call.

“Software now ships with agents in the loop, writing tests, running them, fixing what breaks. A green checkmark was built for a world where a person read every line; it was never built to survive an agent in that loop. Kane CLI carries a requirement all the way to a ship decision and shows its working, a record a teammate, an agent, and an auditor can all act on,” said Mudit Singh, Co-Founder and Head of Growth at TestMu AI.

.evidence is open. The pack is not locked inside a dashboard, it is plain YAML and Markdown, versioned, diffable in git, and readable by people, agents, and auditors. Teams can adopt it without adopting Kane CLI itself, and packs project to CI formats such as CTRF and JUnit. Think of it as the PDF of test evidence.

Source: BUSINESS WIRE