What we're building, in the order we're building it. No dates — we would rather ship a thing than pre-announce it — and everything that lands is written up in the changelog.
The tests your pipeline already runs, reported into the same library. A CI job posts its results and the matching tests get a run each, so a story's coverage field moves when the build does, not only when a person walks the steps. The manual tests and the automated ones sit in one folder tree and one report.
Today the drafted tests are refreshed when somebody presses Refresh. This makes the app notice when a story's description or acceptance criteria change and redraft on its own, telling the model what already exists so it fills the gap rather than starting over. A dismissal still sticks.
The twelve tools already in Rovo and Claude, listed in the other agents people work in — ChatGPT, Codex, Cursor — so a session that ends with a fix can end with the test for it, written into the library against the story, wherever it happened. The tools are served from one Atlassian address for every client, and OpenAI’s directory asks the publisher to prove they own the server’s domain, which only Atlassian can do for its own. That is the step we are waiting on; nothing in the tools themselves changes.
The editor keeps a half-written test for as long as the page is open and no longer. This keeps it in your browser so a closed tab, a crashed one or a click on the wrong link brings you back to the test you were writing, with a note saying when it was last kept.
Shipped in October 2026. Calibro Test is in Claude’s connector directory: the same twelve tools as the Rovo agent, in Claude, Claude Desktop and Claude Code, run as you under your own Jira permissions. A site admin allows external MCP access once in Atlassian Administration; then each person adds the connector and signs in to their site. Read the docs →
Shipped in 1.2.0. Bring existing manual tests in from a CSV or TSV file — map the columns, review every test validated before anything is created, and land them in the project and folder you choose, with their steps and the stories they cover. An import can be stopped and resumed, even after a closed tab, and never creates the same test twice. See the changelog →
Shipped in 1.1.0. The full interface — library, editor, runs, the Tests panel, Coverage and Settings — in German, Spanish, French and Portuguese (Brazil), alongside English. Each person sees the app in their own Atlassian language, no setup needed; any other language falls back to English, and European Portuguese reads the Brazilian catalog first. Your tests are untouched — this is the app's own wording, not your steps or your stories. More languages follow where the requests point. See the changelog →
Shipped in 1.0.0. A library of tests with steps, runs of them one step at a time with a bug filed from the failing step, and a Test coverage field on every story that works in JQL, filters and board swimlanes. Tests drafted from a story by Atlassian-hosted Claude, a Rovo agent, and twelve tools for agents of your own. Runs entirely on Forge with no external egress. See the changelog →