Skip to main content
Post

Build a Jira AI Assistant That Finds Release Blockers

Quickchat published a step-by-step tutorial for building a Jira release-blocker assistant in Quickchat. It's a walkthrough of how to build one, not a new built-in Jira feature.

The example agent, Harbor Release Desk, answers "What is blocking Harbor 1.4?" with a briefing that links each finding to its Jira ticket. It keeps three things separate: tickets labeled release-blocker, tickets with no assignee, and other unfinished work. In the sample project, the login bug depends on a certificate ticket that isn't in the release filter. The assistant follows that blocking link and keeps its direction, so the dependency shows up in the answer.

You build it in steps. Create a test Jira project with two versions and six work items. Then give the agent a narrow main prompt and keep it private by turning off widget and embed access and setting chat page access to Nobody. The first action calls Jira's enhanced JQL search with a fixed project and fix version. The prompt tells the agent to page through results until Jira reports the last page, and to label the briefing incomplete if a page fails. The guide includes the JQL, the prompts, and screenshots.

The first version is read-only. You can then add a second action that drafts a follow-up issue, such as one for the unassigned deployment checklist. It shows the exact summary and project, waits for your explicit confirmation, creates only that draft, and links the new ticket after reading it back from Jira.

You'll need a Jira Cloud test site and a personal API token with the read:jira-work scope. Add write:jira-work if you build the follow-up action. AI replies and Simulation Testing use Quickchat credits. The starter kit holds the configuration, prompt, and sample data. It has no credentials and isn't an automatic import.

Loading replies