Skip to main content
Version: 3.0.0

Frequently asked questions

General usage​

What is the difference between assigning, mentioning, and using Rovo Chat?​

Assigning or mentioning works on the work item. When mentioned, the reply is posted as a comment on the work item; when assigned, the result stays in the agent's panel. Rovo Chat is a private conversation for instructions or feedback without a new assignment. The Rovo Chat conversation is not automatically copied to the work item. See Plan and questions.

What does the agent read?​

The work item summary, description, comments, and text attachments, and its links: repository, branch, pull request, and linked pages. When it plans, answers, or reviews, it also reads the repository in read-only mode. Binary attachments are skipped and the agent says so.

What language does it respond in?​

The work item's language. In Spanish it uses Jira's current terminology: spaces and work items.

Does the agent write to the work item?​

Only when it is mentioned. Its final reply is posted as an agent comment on the work item, replying to the comment that mentioned it. An assignment does not post anything: the result stays in the agent's panel. Failures are not posted either.

Why does it comment when mentioned but not when assigned?​

A mention is a conversation on the work item, so the reply goes back to the same thread. An assignment delegates the work, and its result stays in the agent's panel without adding comments to the work item. See Plan, questions, and lookups.

Does the comment keep its formatting?​

Yes. Tables, code blocks, and lists are displayed with formatting in the comment.

Can I ask something without requesting a change?​

Yes. If the question does not ask for a change, it answers from the work item without opening the repository. If the question is about the code, it reads the repository in read-only mode and answers, without committing or pushing. See Plan, questions, and lookups.

Planning and questions​

Does the plan read the repository?​

Yes, since version 1.1.0. If there is a single repository, before planning it reads the manifest, the package.json files, the AGENTS.md file, and the main entry points in read-only mode. It does not commit or push. If the target is ambiguous, it asks before planning.

What happens if a decision is missing?​

The agent asks and waits for the answer in the chat. Questions are asked only during planning. During implementation there are no questions or permission prompts.

What happens if the site does not have the model or Bitbucket access configured?​

The agent asks an administrator to complete the configuration and does not promise the change. Configuration is performed outside Jira during service setup. See the Administrator Guide.

What data is sent to the model provider?​

The work item text, comments, text attachments, the title and URL of the links, and the repository files it reads. When it implements, it also sends the relevant diff. No secrets or Jira token are sent. See the Administrator Guide.

What does it read from the Teamwork Graph?​

The work item's links: the repository, branch, pull request status, commit subject, and the title and URL of linked pages. It does not download page content or write to the graph.

The Teamwork Graph is an Atlassian feature in preview. If it does not respond, the agent continues with the other sources and says so in the panel.

Which network destinations can the agent's environment reach?​

Only two destinations: the model endpoint and developer.atlassian.com, to read the public Forge documentation. Any other destination is blocked and there is no web search. The Jira token does not enter that environment. See the Administrator Guide.

Does the agent follow Forge's official practices?​

Yes. It uses Atlassian's official guides to build, debug, review, or optimize the cost of Forge apps, depending on what the work item asks for. It does not create, deploy, or install apps.

Pull request, iteration, and limits​

Which repositories does it support?​

A single Bitbucket repository from the configured workspace. If there is no repository or there are several, the agent asks. GitHub URLs are ignored. It does not support multi-repo or multiple Bitbucket workspaces per site.

What is the branch name and how is the pull request created?​

The branch is named deepseek/<key>-<slug>. The pull request is created as a draft and cites the work item key. See Draft pull request.

Does the agent push to the default branch or merge?​

No. The agent does not push to the default branch or merge.

What checks does a change go through before the push?​

Tests, forge lint if the repository has a manifest.yml, and a Forge review. If the change touches permissions, network, authentication, or storage, also a security review. If forge lint fails, there is no push. If the review finds blockers, the agent fixes them once; if they remain, it does not publish. See Draft pull request.

Can it review a pull request that is already open?​

Yes. Mention it with "review this PR". It evaluates the pull request in read-only mode and gives a verdict stating what is wrong and in which file. It does not commit, push, or open another pull request. See Review a pull request.

What happens if implementation fails?​

The message explains why. There is no push to the default branch. If information is missing, the task ends with a suggested question and you can continue with a mention.

How do I iterate on the same pull request?​

Mention the agent with the new instruction. It continues on the same branch and the same pull request, with new commits on that branch. It does not open another pull request or merge.

Can I cancel a task?​

Yes. You can cancel an in-progress task, including during implementation. The work stops and partial work is not published.

What usage limits are there?​

Token, step, and time limits per task. Since version 1.1.0, the default step limit is 80. When a limit is reached, the task stops, the message states which limit was reached and what was completed, and partial work is not published. See Draft pull request and the Administrator Guide.

Installation, version, and scope​

How is the app installed?​

A Jira administrator opens the Forge distribution link provided by MB Group and confirms the permissions. On development sites, the person using the agent must be an app contributor. See the Installation Guide.

What do I need to do when upgrading from 1.0.0?​

An administrator has to accept the new permissions in Apps β†’ Manage apps: write:jira-work, for the comment, and the Teamwork Graph read scopes (read:graph:*), for the work item's links. Without that step, the agent does not comment or read the links. See the Installation Guide.

What happens on uninstall?​

The site's credentials and tasks are deleted. A reinstall creates a new registration and does not mix tasks from the previous cycle.

Is there an administration page inside Jira?​

No. The model, Bitbucket OAuth consumer, and usage limits are configured outside Jira during service setup. To change them, contact the team that manages the service.

Is it available on the Atlassian Marketplace?​

Not in version 1.1.0. Installation uses the Forge distribution link provided by MB Group.

What does version 1.1.0 not include?​

GitHub, Jira Data Center, a Marketplace listing, billing, multi-repo, multiple Bitbucket workspaces per site, and an administration page inside Jira.