The only open-source tool that unifies API test-automation, mocks, performance-testing and UI automation | https://xitter.cf/t.co/zbCi8kGQ8n
Joined February 2017
- Tweets1.2K
- Following10
- Followers3.3K
- Likes1.6K
Karate 2.1.4 is out. JS mistakes now fail as SyntaxErrors with a hint. Credential headers masked in reports by default. Mocks locked down by default. Gatling users no longer capped at CPU x 2.
Step-through debugging: karatelabs.io/intellij and karatelabs.io/vscode
Notes: github.com/karatelabs/karate…
One note on the JS debugging line. The release notes describe clearer JS error reporting in the engine (SyntaxError with hints, the unquoted-variable warning, ASI and TDZ handling), so that is what I have claimed, and I have placed the step-through debugging claim on the plugins, where the product page states it. If Peter had a specific debugger-side change in mind, worth swapping that sentence for his wording.
Karate retweeted
Your policy system never says "I don't know." It returns a premium. A plausible one.
VeriQuant re-rates every case independently, to the cent, and shows the first step where the two calculations part ways.
Bring us the rating case you worry about most. [email protected]
Your policy system never says "I don't know." It returns a premium. A plausible one.
VeriQuant re-rates every case independently, to the cent, and shows the first step where the two calculations part ways.
Bring us the rating case you worry about most. [email protected]
Karate v2.1.3 is out.
This is a stability release, shaped by the community. Most of the fixes came straight from issues our users raised:
- karate.call() in a loop returns fresh responses again
- karate run with karate-pom.json no longer runs every feature twice
- A feature-level @fail tag is inherited by every scenario, as it was in v1
- .xlsx and other Office files read cleanly
- Browser file upload works again on both CDP and W3C
- Error output shows the file name and line number everywhere
- Mock server hardening: all request data is treated as untrusted
On top of that: JS engine conformance work, Http.quiet() for tests that expect a 4xx or 5xx, better run events for listeners, and current netty, jackson and logback.
Release notes: github.com/karatelabs/karate…
Thank you to everyone who filed an issue or sent a pull request. Open source stays sharp because you keep pushing it.
If Karate already runs in your pipelines, there is more on offer for enterprise teams: IDE tooling, Karate Agent for AI-driven test creation, and support built for regulated environments. This short video covers it: youtube.com/watch?v=NEcC6Pu0…
When the same AI writes the code and the tests that check it, green is self-graded.
Karate Agent draws a hard line: AI proposes, deterministic code decides. No model in the path. Runs inside your walls.
Know what's safe to ship. Even when AI wrote it.
Contact us for a demo karatelabs.io/contact-us?per…
1/ Your coverage is green. Would your tests notice if the code broke?
Different questions. The gap between them is where incidents live.
6/ Echo assertions cannot inflate the number. The independent score counts only assertions with no detected dependence on the server's answer. A suite of echoes scores zero.
7/ Scope, stated plainly: it grades assertion strength, not correctness. And as far as we can find, no shipped tool does this without the system's source.
Start with the suite you already run and a base URL.
How it works: karatelabs.io/api-mutation-t…
Blog: karatelabs.io/blog/would-you…
Every task ticked. Build green. Ready to ship?
AWS Kiro is the most serious spec-driven development tool out there: requirements, design, and tasks written down before the code exists.
But like every spec-driven tool, it records completion the only way it can. An agent ticks a checkbox in tasks.md. Nothing checks whether any test exercised the acceptance criteria that task named.
We built the missing layer and put a live example on the public internet:
12 of 12 scenarios pass. 100% API operation coverage. Every task ticked. Release verdict: NOT READY.
One criterion, APR tier boundaries, is marked done by task 5 and exercised by no test. Invisible to the dashboard. Invisible to the agent that wrote the code.
Karate reads .kiro/specs in place, joins every acceptance criterion to real test runs, and computes a verdict a checkbox cannot influence. Install the Karate Power and Kiro asks for the verdict itself over MCP, then writes the missing check.
Self-hosted. Bring your own LLM. The report is public and needs no install: karatelabs.io/kiro
1. Keep Playwright. Take the deck.
We just shipped test data where every row carries the computed expected result. Plain JSON/CSV. Your framework runs it. Zero Karate in the loop.
6. Run the deck from Playwright, Postman, or REST-assured as data. Run it inside Karate and each row becomes evidence against the requirement it names: traceability matrix, coverage grade, release verdict.
The deck is the on-ramp.
