Rovo Agent: Unable to Reliably Remove Duplicate Entries in Generated Report

[edrick_deguzman]

I’m running into a persistent issue with deduplication in a Rovo agent that generates a daily Jira update report, and I’m trying to understand whether this is a platform limitation or something I’m missing.

Use case:

• The agent generates a daily report based on tagged Jira comments created in the last 24 hours
• Tags include: [DONE], [IN-PROGRESS], [BLOCKED], [ACTION-REQUIRED], [NEXT]
• The report has multiple views (Leadership View, Highlights, Team View), but all views are supposed to operate only on a pre-selected set of comments

Key requirement
For each Jira issue and each comment author:

• If the same author adds multiple tagged comments on the same issue within the 24-hour window
• The report must include only the most recent tagged comment from that author for that issue
• All earlier tagged comments from that author on that issue should be excluded entirely

Other authors’ comments on the same issue should still appear normally.

What I’ve already implemented
In the agent prompt, I’ve explicitly instructed the AI to:

• Select exactly one tagged comment per (issue × author)
• Use the most recent comment by creation timestamp (UTC)
• Treat earlier comments as discarded / invalid, not just “deduplicated”
• Prevent views from re-selecting or re-introducing additional comments
• Ensure views only operate on the already-selected comments

Despite this, I still occasionally see output like: 

Author1
- [BLOCKED] CNSG-1251: Integration blocked pending API endpoint
- [IN-PROGRESS] CNSG-1251: API and design received, integration started

Both comments:

• Are from the same author
• Are on the same Jira issue
• Fall within the same 24-hour window
• But only the latest ([IN-PROGRESS]) should appear

[Dr Valeri Colon _Connect Centric_ · Accepted solution]

Welcome to the community @edrick_deguzman Rovo can’t reliably enforce per-author, per-issue deduplication because it doesn’t build a structured dataset before reasoning. It sees comments as text, not stable records, so it may re-include older items even with strict instructions. This is a platform limit, not your prompt. The only reliable solution is to prefilter the comments outside Rovo—via Automation, Forge, or an MCP script—and pass Rovo only the final list.