Inspiration

I was the only frontend developer on a 9-person team. A lot of my waiting looked like this: I asked the backend developer a question in a Jira comment ("@backend, does this endpoint return 409 or 422?"), the mention notification got buried, and the issue sat until someone remembered. The person who asks remembers the question. The person who was asked usually does not. Jira shows comments, but nothing shows which questions are still unanswered.

What it does

Left on Read finds questions in Jira comments and tracks whether the person they were addressed to has replied.

  • Issue panel: open it from the app button on a work item. It lists each question in that issue's comments: who asked whom, and how long it has gone without a reply. Answered questions move to a separate list with the reply time.
  • Global page (Apps > Left on Read): questions on work items updated in the last 60 days, in three tabs: Waiting on me, I'm waiting on, All open. The header shows the longest current wait and the average reply time for answered questions.
  • Mark answered: when the answer came in chat or in a meeting, anyone can close the question by hand. It can be reopened from the issue panel.
  • Rovo action and skill: the list-unanswered-questions action returns open questions as text (inputs: scope = waiting-on-me / asked-by-me / all, optional projectKey). A rovo:skill with a SKILL.md tells an agent when to call it.

A comment counts as a question when it mentions someone and contains a question mark. It counts as answered when the mentioned person writes any later comment on the same issue, or when someone marks it answered. Mentioning yourself is ignored, and a comment with two mentions creates two questions. A comment like "@backend FYI, merged" is not a question.

How I built it

  • Forge UI Kit (@forge/react) with one resource shared by jira:issuePanel and jira:globalPage.
  • A resolver reads data with api.asUser(), so users only see questions on issues they can already see: POST /rest/api/3/search/jql for recently updated work items and /rest/api/3/issue/{key}/comment for comments.
  • Comment bodies are Atlassian Document Format. The detection walks the ADF tree to collect mention nodes and text. This logic sits in src/questions.js with no Forge imports, so I could test it with plain Node before deploying.
  • Manual "answered" marks are stored in Forge KVS per issue.
  • The Rovo action and rovo:skill modules reuse the same functions.
  • Scopes: read:jira-work, read:jira-user, storage:app. No external egress.

Challenges

  • The first manifest used a rovo:action module name; the linter rejected it, and the current schema uses action plus rovo:skill.
  • Comment text alone is not enough, because mentions are separate ADF nodes. Walking the tree was simpler and more reliable than parsing rendered HTML.
  • To show a real reply in the demo I used a second account on the dev site and posted the reply as that user.

What is not done yet

  • Rovo is not enabled on my developer site, so the video shows the action and skill in the manifest but not an agent calling them.
  • The global page checks the 50 most recently updated work items from the last 60 days, and up to 100 comments per work item.
  • Replies outside Jira are not detected; that is what Mark answered is for.

What's next

Optional reminders to the mentioned person after a wait threshold, and per-project settings for what counts as a question.

Testing instructions for judges

  1. Install with the installation link (Jira). No configuration is needed.
  2. In any project, add a comment that mentions a teammate and contains a question, for example @Alex is the API returning 409 here?
  3. Open the work item, click the app button under the title and choose Left on Read. The question appears with "no reply for N min".
  4. As the mentioned teammate, add any comment on the same work item. Reload the panel: the question moves to Answered.
  5. Open Apps > Left on Read for the site-wide tabs, and try Mark answered on an open question.
  6. If Rovo is enabled on your site, ask an agent "what questions are waiting on me?".

Code (MIT): https://github.com/kyunghoonkook/left-on-read

Built With

Share this project:

Updates

Submission history