Inspiration

Support teams triage the same way every day: read a ticket, guess a category, set a priority, find the right person. It is repetitive, slow, and inconsistent — and every minute spent triaging is a minute not spent helping the customer.

What it does

Support Triage Copilot is a Forge app that adds a panel to any Jira issue. It suggests a category with a confidence score and the matched keywords, a priority derived from urgency signals, and the best assignee from an editable team roster. It searches Confluence for help articles matching the ticket, and applies the category label, priority, and assignee in one click — adding a comment that records exactly what changed. If the suggestion is wrong, the agent corrects it and the app stores the correction as keywords, so future suggestions improve. Everything is rule-based: no LLM calls, no external servers, no black box.

How we built it

Atlassian Forge with UI Kit (@forge/react) for the issue panel, a Forge resolver backed by Forge key-value storage for the roster and learned keywords, and Jira + Confluence product APIs through @forge/api. Built and deployed with the Forge CLI (Node.js).

Challenges we ran into

The current @forge/api requires product API URLs to be built with the route export — plain-string URLs are rejected at runtime. The panel first shipped with that bug, errored on load, and was fixed by converting every call to route templates. The panel's suggestions also had to be explainable, so every suggestion shows its evidence: matched keywords, scores, and urgency signals.

Accomplishments that we're proud of

Verified end to end on a live Jira site: two real test issues triaged (a billing double-charge and a locked-out access request), production deployment, and a public installation link any Jira site can use.

What's next

More categories and urgency signals, suggesting a KB article draft when none matches, and balancing assignments across the team's open workload.

Share this project:

Updates

Submission history