Aggregation Method Implementation using Instruction Prompts - Rovo Studio is not working at all

Ranvijay Singh

I'm looking for guidance from the Rovo Studio community on implementing numeric aggregation using instruction prompts.

My agent successfully retrieves Jira issues using JQL, and the issue count consistently matches the Jira Board. However, aggregating Story Points (SUM) is unreliable—the total is intermittently incorrect even though all records appear to be returned.

I've tried explicit instructions to iterate through every record, maintain a running total, validate processed records, and reconcile with the board, but the results are still inconsistent.

Has anyone successfully implemented reliable aggregation (SUM, AVG, COUNT, etc.) using instruction prompts alone? Or is this a current limitation that requires a different approach or tool? Any best practices or examples would be greatly appreciated.

Thanks,

Ranvijay Singh

Josh

Hi Ranvijay, in my experience this is a current limitation rather than a prompt problem. The agent can retrieve the right records, but it adds them up as a language model, token by token. It doesn't run a real SUM. Instructions like "keep a running total" make it look more careful without making it reliable, especially once you have more than a couple of dozen records.

The approach that holds up is: compute in Jira, let Rovo explain.

• Automation (no code): Scheduled or manual trigger, then Lookup work items with your JQL. Then use smart value math: {{lookupIssues.size}} for COUNT, and {{lookupIssues.Story Points.sum}} for SUM. .average, .min and .max work the same way. You can use the field's custom field id instead, e.g. {{lookupIssues.customfield_10016.sum}}; check which story points field your board actually uses. Pass those numbers into the Use Rovo agent action as input, or write them to a comment or page. Lookup is capped at 100 items per call, so split by sprint or team if you're above that.
• Beyond 100 items: a small Forge action exposed to your agent, which calls the Jira search API, pages through everything and returns the totals, gives exact numbers at any size. The agent just calls it.
• Whatever you choose, have the agent output the key + value list alongside the total, so anyone can spot-check it against the board.