Inspiration Every Jira board quietly rots: issues stall without updates, due dates slip, blockers sit unescalated, and nobody owns the cleanup. Managers discover it in the weekly meeting — too late. We wanted an agent that watches the board continuously, explains risks in plain language with evidence, and doesn't stop at reporting: it fixes what it finds.

What it does FlowPilot is a Rovo agent for Jira that audits a project board on demand:

  • Computes a board health score (0–100) from deterministic, explainable rules — no black-box scoring.
  • Flags every at-risk issue with concrete reasons: overdue (with days), blocked (label or an open "blocks" link), unassigned, stale (no updates beyond a threshold), missing due date, and long-running in-progress work.
  • Suggests the single most useful fix per issue and, with the user's confirmation, executes it: assign an owner, set a due date, post an escalation comment, or publish a full audit report as a Confluence page.
  • A Jira global-page dashboard shows the same audit without chat: pick a project and see its health score, per-risk-type counts, and the top at-risk issues in a table — the numbers always match the agent's audit because both run the same engine.
  • The same five actions are also exposed as Atlassian MCP tools, so teams can drop FlowPilot's abilities into their own custom agents in Rovo Studio.

How we built it

  • Forge (Node.js 22) with a rovo:agent module, five action modules, a rovo:mcp module reusing the same actions as tools, and a jira:globalPage dashboard (UI Kit) backed by the same audit engine.
  • A pure, deterministic audit engine (weighted risk scoring) with unit tests — the agent's claims are always backed by reproducible computation, not LLM guesswork.
  • Jira REST (paged /search/jql) and Confluence v2 APIs via @forge/api, running as the app so results are consistent for every user.
  • Demo data engineered on the dev site so each risk type is visible at a glance (overdue / blocked / unassigned / healthy control issues).

Challenges we ran into

  • Forge manifest validation for Rovo + MCP modules is strict (description length caps, MCP module shape) — solved by iterating with forge lint.
  • Making an agent's output trustworthy: we moved all detection into deterministic code and let the LLM layer only explain and converse.

Accomplishments that we're proud of

  • End-to-end working agent: audit → explain → remediate → publish, verified against live Jira/Confluence data.
  • One audit engine, three surfaces: Rovo agent, MCP tools, and a Jira dashboard — on the prepared DESIGN demo board all three agree (health 63/100, 9 at-risk issues).
  • One action set, two surfaces: Rovo agent and MCP tools.

What we learned Deterministic scoring plus an LLM that only explains turns an agent from a demo into something a team can trust: every number in the chat can be reproduced by the dashboard and by unit tests.

What's next for FlowPilot

  • Health trends over time on the dashboard, from scheduled snapshots.
  • Scheduled autonomous audits with Teamwork Graph context (who is overloaded, which team owns the blocker).
  • Slack/Teams digests of the top risks each morning.

Try it out Install FlowPilot on your own Atlassian site (Forge distribution link): https://developer.atlassian.com/console/install/988b7bb5-1354-4cc1-b46b-2563e02d0ba0?signature=AYABeL8qm%2BXJYmR2YlsNQApySwkAAAADAAdhd3Mta21zAEthcm46YXdzOmttczp1cy13ZXN0LTI6NzA5NTg3ODM1MjQzOmtleS83MDVlZDY3MC1mNTdjLTQxYjUtOWY5Yi1lM2YyZGNjMTQ2ZTcAuAECAQB4CxqLTo9l4lqXuFjNim1xJM%2Bp%2F3TVP9xHS2BrU5DRRucBqMH62LJyWhdgtZcwtjEqdgAAAH4wfAYJKoZIhvcNAQcGoG8wbQIBADBoBgkqhkiG9w0BBwEwHgYJYIZIAWUDBAEuMBEEDE2P%2FWUnIhG8kNUBRgIBEIA7C6aIwSyLgnCJnJH5rPul%2BDMbYc%2Bgi%2Fwpt0yBG2h4JcRUgigjFWoJaM39%2FKIkpOZLCyoFZRzRZx42rFMAB2F3cy1rbXMAS2Fybjphd3M6a21zOmV1LXdlc3QtMTo3MDk1ODc4MzUyNDM6a2V5LzQ2MzBjZTZiLTAwYzMtNGRlMi04NzdiLTYyN2UyMDYwZTVjYwC4AQICAHi9bVMAkme4qVQX1fQHo6ePw3H40A7cGfz7yrmSfBWMXgEdOenNFtThv%2BzL0v%2B1zOLRAAAAfjB8BgkqhkiG9w0BBwagbzBtAgEAMGgGCSqGSIb3DQEHATAeBglghkgBZQMEAS4wEQQMrpyyY51od5FKB%2FxnAgEQgDsNk3neAUi%2B%2BtDQKl%2BA7mc0s%2FrEzZf%2B%2FXmNry1PKcyC%2BneCPpw1q5wOBGoSXwiU6RtO5Gc2UTvqH5%2FdDQAHYXdzLWttcwBLYXJuOmF3czprbXM6dXMtZWFzdC0xOjcwOTU4NzgzNTI0MzprZXkvNmMxMjBiYTAtNGNkNS00OTg1LWI4MmUtNDBhMDQ5NTJjYzU3ALgBAgIAeDYASx0XCjkny4iLV6vz%2BKtEGSGT6kdt0QRpkn6D%2FlVEAa3CXc1lScz9X0JbNDAxfQEAAAB%2BMHwGCSqGSIb3DQEHBqBvMG0CAQAwaAYJKoZIhvcNAQcBMB4GCWCGSAFlAwQBLjARBAx4YcB6PeQOEalRKPwCARCAO3lmZ3GPQJnKTQB2WMKS0PkgajeTOI7nEQDKverM1XZuHWjSymzYJyb9trQp1oUk11sy4KuDOO0lyAyGAgAAAAAMAAAQAAAAAAAAAAAAAAAAAKjiN%2BhbsSuXy8A6kVD0GB%2F%2F%2F%2F%2F%2FAAAAAQAAAAAAAAAAAAAAAQAAADIeMz%2FRCWWqibwxjfQGSjCW6kV8%2BwxGJKDHzjLrbs7hHoaEKlrKcWtQR636J1lQJuSq1Tf2SYfWLc03LOUEmN%2FzzzM%3D&product=confluence&product=jira Or explore the prepared demo site https://one-atlas-ungo.atlassian.net — project DESIGN is the prepared demo board (health 63/100, 9 at-risk issues).

Share this project:

Updates

Submission history