Introducing coverage gap detection: How the Zephyr Agent for Rovo ends duplicate test cases 

AI has changed how fast testing teams can move, yet sorting through duplicate test cases still trips teams up. Ask the SmartBear Zephyr Agent for Rovo to generate test cases for a requirement, and coverage that used to take an afternoon comes back in a single conversation. That speed is a genuine win for teams building faster than ever, but as AI generates more test coverage, how do you make sure all of it stays worth keeping? 

With Zephyr 15.4, SmartBear has launched coverage gap detection for the Zephyr Agent for Rovo, which makes the agent smarter about exactly this. Every test case it generates now fills a real gap, checked against what already exists and linked to the work item, so your team gets the full benefit of AI-speed testing without any of the cleanup that used to come with it. 

New to the Zephyr Agent for Rovo? Read the full explainer:  

What Is the SmartBear Zephyr Agent for Rovo? AI Testing in Jira, Explained 

Key takeaways 

• Zephyr 15.4 launches coverage gap detection for the Zephyr Agent for Rovo: it reviews what already exists and is linked to a work item, then generates new test cases only for the scenarios that aren’t covered yet, no extra step required. 
• Your team gets the test cases it’s missing, and the time that used to go into catching duplicates goes back to real testing work. 
• Want a deeper search? You can run the check yourself and have the agent scan your whole test case library to find related test cases before creating anything new. 
• The same release also adds test case and test plan search directly from Rovo chat, and a permission check before generation, so access issues surface immediately, instead of after the work’s already done. 

What duplicate test cases actually cost your team 

A work item that gets revisited across a few sprints can collect five, six, or more overlapping test cases, each one written by someone who didn’t know the others existed. Someone eventually has to catch these duplicates, and catching the error means reading through what’s there, comparing it to what’s new, and deciding what to keep, archive, or merge. That’s real time, and it’s time spent on cleanup instead of on testing what actually still needs coverage. 

It also has a subtler cost: once a test library has enough duplicate noise in it, people stop fully trusting what it says. If Zephyr is your system of record for testing, that record only works if the team believes it. That trust is the assurance everything else rests on, and the clutter is what erodes that belief first. 

This is why we prioritize delivering application integrity: continuous, measurable assurance that your software just works as intended, with governance to operate at AI speed and scale. Deduplication is one small, concrete piece of it – governance to maintain pace with AI-driven testing and sort through the noise. 

With Zephyr, spend your time testing, not deduping  

As of Zephyr 15.4, the agent handles this automatically. When you ask the Zephyr Agent for Rovo to create test cases for a Jira work item, it checks whether test cases are already linked to that item, reviews what those existing test cases cover, and creates new test cases only for the scenarios that aren’t covered yet. You don’t have to ask it to check first, and nobody on your team has to comb through the results afterward looking for repeats. It’s part of how test case creation works now, and it’s testing made easier by design. 

The outcome is simple: less time spent maintaining the test library, and more confidence that the coverage Zephyr shows you is the coverage you actually have. 

Want more control? Scan the library before creating new tests. 
The automatic check looks at what’s already linked to the issue, which covers the common case well. But linked test cases aren’t the whole picture: if a test case that already covers a scenario was never linked to this particular issue, the automatic check won’t catch it. For a deeper look, one that searches the entire test case library rather than just what’s already connected, the same idea is available as a longer sequence of prompts. Follow the steps below to execute this process. 

Step 1: Fetch the issue and pull search terms. 

Prompt: Fetch the details for PROJ-123. Break the summary, description, and acceptance criteria down into individual nouns, verbs, and attributes that could work as test case search terms. 

This returns the issue’s summary, description, and acceptance criteria, plus a list of extracted terms, for example: login, email, invalid, session, and timeout. 

Step 2: Check what’s already linked. 

Prompt: What test cases are already linked to PROJ-123? 

This returns the linked test case keys, for example PROJ-T45 and PROJ-T46, so you can  exclude them from whatever gets proposed later. 

Step 3: Search the test case library using the extracted terms. 

Prompt: Use each of the terms from Step 1 as a separate search in the test case library. 

Rovo runs one search per term and returns a numbered list of matching test case keys and names for each. Since matching happens on test case name and key text only, broad, generic terms turn up more candidates than exact phrases copied straight from the issue. 

Step 4: Fetch full details for the matches.

Prompt: Get the full details, including objective and test script, for the test cases you found. 

Only name specific keys if you want to narrow it down to particular ones, otherwise Rovo pulls full details for everything that matched. Either way, this gives you the objective, precondition, and script for each candidate, which is what the agent needs to judge fit rather than just going on the name. 

Step 5: Ask the agent to filter and justify.

Prompt: Compare these test cases against PROJ-123's summary, description, and acceptance criteria. Exclude PROJ-T45 and PROJ-T46, they’re already linked. Propose only the ones that are genuinely relevant, and explain why each one fits. 

You get back a shortlist of proposed test cases, each with a short explanation of why it fits. 
Prefer to run the whole thing in one prompt? State it all up front and let the agent work through the steps in sequence: 

Prompt: Fetch the details for PROJ-123. Deconstruct the summary, description, and acceptance criteria into a list of every individual noun, verb, and attribute mentioned. Use each of these as a separate, broad search term in the Zephyr test case library to maximize coverage. For all matches found, fetch their full test case details, including objectives and test scripts, to evaluate their actual relevance. Propose only the most relevant test cases that are not already linked to the issue, explaining why each one fits the criteria. 

Tips for getting the most out of the dedupe capability 

• This only works from the Zephyr Agent for Rovo, not generic Rovo chat. Make sure Zephyr Agent for Rovo is selected before running these prompts; Zephyr’s skills aren’t available otherwise. 

• Favor broad, generic search terms over exact phrases pulled from the issue. Matching happens on test case name and key text only, so a single word like “checkout” surfaces more than a specific phrase would. 

• Working with an issue that produces a long list of search terms, or a search that turns up a lot of matches? Ask for just the candidate list first, then only pull full details, and a relevance check, on the ones worth a closer look. 

Also launching in this release:  

• Search from Rovo chat: Find the right test case just by asking. Zephyr 15.4 also brings your whole test library into Rovo chat. Give the agent a search term, and it matches that term against test case and test plan names and keys, so finding the right test case is as easy as asking for it. Today, this searches names and keys. 

• A permission check before generation: Catch access problems before you do the work. The Zephyr Agent for Rovo now checks whether you actually have permission to create test cases in a project before it generates anything. If access is the problem, you find out immediately, instead of the agent doing the work first and only then telling you it couldn’t save the result. 

The impact: A test library that scales alongside AI 

This process, using the Zephyr Agent for Rovo, reduces how much time is spent maintaining testing records, and how much confidence you can place in what that record says. As a result, you have less duplicate cleanup, less second-guessing, and a test library that only gets more valuable as your team leans on AI to generate more coverage, faster, without generating more noise to manage. 

If you’re already on Zephyr v15, deduplication is live for you right now, no setup required. If you don’t see it, or don’t see the Zephyr Agent for Rovo at all, upgrading to the latest version of Zephyr activates it immediately, alongside every other improvement we’ve shipped since your last update. 

This is what testing at AI speed should feel like: faster coverage, cleaner results, and an agent that gets smarter with every release and keeps your test record trustworthy as fast as AI can fill it. Try it on your next work item and see the difference for yourself. 

New to Zephyr? Start your free trial on the Atlassian Marketplace and see deduplication in action from day one. 

Frequently asked questions 

• How does the Zephyr Agent for Rovo avoid creating duplicate test cases? 

When you ask it to create test cases for a work item, the Zephyr Agent for Rovo first checks what’s already linked to that item, reviews what those tests cover, and then writes new cases only for the gaps. The process is built into the test case creation workflow: you don’t trigger the check, and nobody has to comb through the results for repeats afterward. 

• Do I need to do anything to turn on test case deduplication in Zephyr?  

No. If you’re on Zephyr v15 or later, deduplication runs automatically whenever the Zephyr Agent for Rovo creates test cases. 

• Can I still choose BDD versus step-by-step format?  

Yes. The automatic check picks up whatever format you ask for; running the manual prompt sequence just gives you visibility into the coverage gap before anything gets created. 

• Does the chat search look inside test case steps?  

Not yet. Today it matches search terms against test case and test plan names and keys.