Blast Radius

Inspiration

Every engineer has shipped a "small" change that broke something big. The warning signs usually existed (a past incident on the same service, a teammate editing the same code, an epic due next week), but they were scattered across Jira and nobody connected them in time. Blast Radius connects them before you ship.

What it does

Blast Radius scores any Jira work item from 0 to 100 and explains why: related past incidents, open incidents, work that depends on the change, parallel changes, and at-risk epics. It also tells you who to loop in.

It works in three places:

• a Jira issue panel, which can post the report as a comment
• a Rovo agent you can ask "What's the blast radius of PAY-2?"
• a Forge MCP server, so tools like Claude Code can call it

How I built it

One Forge app with an issue panel, a Rovo agent, four actions, and an MCP server. It reads Jira and JSM data, and walks the Teamwork Graph with persisted Cypher queries. Each incident gets a relevance score:

$$ r_i = \text{severity}_i \cdot e^{-\text{age}_i/120} \cdot \text{overlap}_i $$

and the final score is a sum of capped, named factors, so every point is explained:

$$ S = \min\left(100,\ \sum_k \min(\text{cap}_k,\ \text{factor}_k)\right) $$

Challenges

• Forge apps in Bitbucket can't call Jira, so I rebuilt it as a Jira app.
• Teamwork Graph only runs pre-registered queries, and I found the right field names through trial and error.
• A few platform quirks along the way: module imports, link direction, and dev-vs-prod differences.

What I learned

The hard part isn't the AI. It's gathering the right context. A simple, transparent score that people can check builds more trust than a black box.

Built With

• atlassian-forge
• atlassian-rest-api
• claude-code
• cypher
• elevenlabs
• ffmpeg
• forge-ui-kit
• graphql
• javascript
• jira
• jira-service-management
• mcp
• node.js
• playwright
• react
• rovo
• rovo-agent
• teamwork-graph
• vitest