A practical guide to Atlassian Rovo Agent Skills in 2026

TL;DR

Rovo Agent Skills are reusable, multi-step instructions that let Atlassian's AI agents do real work in Jira, Confluence, and Jira Service Management, not just answer questions. Atlassian now splits them from tools, the single actions a skill calls, like "Create work item". They cluster into three jobs: creating content, automating tickets and projects, and finding or analysing information. Since September 2026 you can write custom skills in plain language in Rovo Chat or Rovo Studio, and developers can add custom tools with Atlassian Forge.

The 2026 catch: Rovo is included on paid Standard, Premium, and Enterprise Cloud plans and metered in Rovo credits, with usage over the allowance billed at $0.01 per credit from December 3, 2026. Its deepest actions stay inside Jira and Confluence, and you test an agent with sample prompts rather than a replay of your past tickets.

I build AI agents for a living, and after years of putting them on live support queues I've learned the hard part isn't the skill, it's the knowledge and the testing around it. That's why I'd reach for a cross-platform option like eesel AI, which connects Confluence, Slack, your help desk and more, then lets you simulate the agent on your past tickets before it ever replies to a real person.

What are Atlassian Rovo Agent Skills?

Rovo Agent Skills are reusable, multi-step instructions that let a Rovo AI agent do more than just answer questions. Atlassian now separates skills from tools: a tool is one action, like "Search with JQL", and a skill strings tools and instructions together, like "Draft release notes" or "Summarize incident". Together they’re what turn the agent from a simple search tool into an active AI assistant that can take action in your workflow.

These skills cover a bunch of different actions, like creating Jira issues, publishing Confluence pages, or acting in a few third-party apps, such as sending a Slack message or an email from Gmail once you connect your account. The whole idea is to give an agent a focused set of skills for a specific job. In fact, Atlassian recommends no more than five tools per agent to keep performance high (its tools guide has the full list). By limiting what an agent can do, you can build specialized helpers, like a "Release Notes Drafter" or a "Bug Triage Assistant," that are really good at one thing. If you're weighing the wider toolset, my roundup of AI agents for Jira is a good companion read.

What can you actually do with Rovo Agent Skills?

Atlassian ships built-in skills such as Write PRD, Write status update, Draft release notes and Summarize incident, and they mostly fit into a few main categories that help teams with common, everyday headaches.

Creating and managing content

Nobody enjoys drafting documents from scratch. These skills are all about getting an agent to do the heavy lifting of generating and organizing content within your Atlassian tools.

For example, an agent can scan all the completed Jira issues in a release and automatically pull together a draft of the release notes on a Confluence page. Or, if you have a long meeting transcript, you can ask an agent to create a quick summary with the key decisions and who’s responsible for what. It can also spin up new project pages from templates, the same idea behind onboarding flows in Confluence, saving you a few clicks every time you start something new. This is the same territory as Atlassian Intelligence in Confluence, just packaged as agent actions.

Automating tasks and projects

This is where Rovo agents start to feel like a real helping hand, directly managing your Jira issues and helping to clean up messy backlogs.

• Tidying up tickets: An agent can look at new bug reports and check if they have all the necessary info, like steps to reproduce the issue. If something’s missing, it can flag the ticket or add a comment asking for more details.

• Organizing the backlog: A product manager could ask an agent to find all the unassigned issues with the "Q3-feature" label and pull them into the current sprint.

• Updating in bulk: You can tell an agent to comment on or change the status of up to 20 Jira tickets at once. It’s a small thing, but super handy for clearing out old, stale issues. Teams running Jira Service Management AI lean on this for routine queue hygiene.

Finding and analyzing information

These skills help you dig up information that’s buried deep inside your Atlassian account, turning scattered data into something you can actually use.

In Jira Service Management, for instance, an agent can find other tickets with similar descriptions. This is great for spotting recurring problems before they become a huge deal. You could also feed it a list of user stories, and it can suggest how to group them into logical epics based on common themes. For a fuller picture of where the platform is headed, see my review of Jira Service Management's AI.

How do you build your own Rovo Agent Skills?

You describe a custom skill in plain language in Rovo Chat or Studio, or a developer builds custom tools in Forge. Atlassian gives you two paths for building custom agents, and they’re aimed at very different people. This is where things can get a little complicated.

First, there’s Rovo Studio, the low-code/no-code approach. This is a visual builder designed for people who aren’t developers. The process is pretty simple: you tell the agent what its goal is, point it to knowledge sources like specific Confluence spaces or Jira projects, write instructions in plain English, and pick from a list of pre-made actions. Since the September 2026 update you can also write custom skills there or in Chat by describing the process, and Studio's automation flows add triggers, branches and conditions. It’s good for well-defined tasks, but anything beyond Atlassian's built-in tools, MCP connections or Marketplace tools needs a developer.

Then you have Atlassian Forge, the pro-code approach. This is Atlassian's full developer platform, which gives you all the flexibility you could want for creating custom apps and agents. It requires you to write JavaScript or TypeScript to define your own logic and actions, and Forge tool modules you publish can be used by Rovo agents in your org or on the Marketplace. While it's powerful, it also means you need a developer on hand, which is a big ask for most support, IT, and operations teams. If a developer route isn't realistic for you, an AI plugin for Jira that's configurable without code is usually the saner middle ground.

What are the limits of Rovo Agent Skills?

Rovo skills work best inside Jira and Confluence: they can read many outside apps, but their actions there are narrower, testing uses sample prompts, and every run spends credits. Here are a few real-world problems you’ll likely run into.

• Your knowledge is everywhere: Most teams don't keep everything in Atlassian. You probably have critical information in Google Docs, important conversations happening in Slack, and your customer support history living in a help desk like Zendesk. Rovo can read much of it through connectors, but its actions in those tools are limited to things like sending a message, and the third-party actions Atlassian lists are things like sending a message, not working tickets in a helpdesk.

• The "too simple" or "too complex" problem: Custom skills and automation flows in Studio have narrowed the gap, but anything beyond the available tools still means Forge, which is too technical for most business users. Agents also get shaky as you pile on skills: one Reddit user found that agents struggle "with too many skills or too complex instructions."

• Testing runs on examples you write: Rovo's builder has a testing pane where you type sample queries and check the answers, rather than a dry run over your own past tickets. Testing only on examples you think of feels risky, and it's a fast route to the kind of AI hallucinations that erode trust in a support context.

If your knowledge and tickets live outside Atlassian, it's worth looking at tools designed to work across them from day one, like eesel AI.

Where do Rovo Agent Skills fall short, and when should you use eesel instead?

Rovo skills are a good fit when the work is already in Atlassian. They're built into Jira and Confluence, they only use what each person can already see, search and summaries are free, and paid seats come with credits. For a team that lives entirely in Atlassian and mostly wants search, summaries and Jira housekeeping it already pays for, Rovo is still the better pick.

The limits are the ones above: actions centre on Jira and Confluence, testing uses sample prompts, and every agent run spends credits that are billed at $0.01 each over the allowance from December 3, 2026 (Atlassian's credit rules).

eesel AI fits when support runs across more tools. It's an AI teammate that works inside Jira Service Management through its Jira integration and reads docs through its Confluence integration (read-only), alongside Google Docs, Slack, Microsoft Teams and helpdesks like Zendesk and Freshdesk.

It acts, not just searches: it creates, assigns and transitions Jira issues, updates fields, replies to the customer or adds an internal note, all set up in plain language without code. Grounding it takes minutes, as my notes on training AI on your knowledge base show, much like a good AI for IT service management setup.

You can test it on real questions before it goes live and see its likely resolution rate. Rovo has Slack and Teams apps too, but everyone who asks it something needs an Atlassian account on a Rovo-enabled site, and replies in public channels need a shared account an admin sets up. With eesel, the person asking needs no Atlassian seat. One difference to know: eesel reads Confluence through one Atlassian API token, so it sees what that account can see rather than each asker's own permissions. Pricing is per ticket or chat, not per Atlassian seat. For more options, see my Rovo alternatives guide.

How much do Rovo Agent Skills cost?

• Atlassian Rovo: As of 2026, Atlassian Rovo is auto-included in the paid Standard, Premium, and Enterprise Cloud plans, and usage is metered in Rovo credits with a monthly allowance per plan. That allowance is 25, 70 or 150 credits per user a month on Jira or Confluence (250, 700 or 1,500 on the collections), and a basic agent run costs 10 credits. There's no separate Rovo line item, but heavier agent use can push you past the allowance, where extra usage is billed at $0.01 per credit from December 3, 2026, on by default, so the "free" framing only holds until you actually lean on it. (Atlassian also sells a separate Rovo Dev agent for engineers at $20 per developer per month, though Rovo Dev as a standalone product is heading for end of life.)

• eesel AI: eesel AI has straightforward credit pricing with no per-seat fees and no hidden charge for each resolution. You pick one fixed monthly price sized to your volume (from $299 a month for 500 credits, where 1 credit is one ticket or chat), with a free 100-credit plan to start and the option to move up or down any time, so the cost scales with value instead of being baked into a bigger subscription you're already paying for.

The bottom line: Think beyond a single platform

Rovo Agent Skills are a decent step forward for automation, and they pay off most when your team's work lives inside Atlassian products. For most of us, that's just not how work gets done.

Real workflow automation needs a tool that can connect to and act on information wherever it lives, across your help desk, internal wikis, and team chat. If you're running support or IT and want something that's easy to set up, powerful enough for real-world workflows, and works with your entire tech stack, a dedicated platform like eesel AI is probably the more practical choice, and my guide to the IT help desk in 2026 digs into where that pays off.