Attention items: from alert to coding task

Definition

Not every alert deserves a code change. This is DevFlow's holding place for signals from outside: something happened, and someone, or some agent, needs to decide what it means. It belongs to the project-management surface, next to stories and labels.

Where they come from

The Sentry integration polls a project's unresolved issues every 2 minutes. You configure it with the organization slug, the project slug and an auth token, plus a base URL for a self-hosted Sentry server. Each new issue becomes an entry, and the recurring dependency scan adds its findings the same way.

Triage and analysis

A triage role looks at each new error entry first, with a 2-minute timeout. Entries rated critical or high go on to a deeper analysis with a 10-minute timeout, which needs at least one repository in the workspace. The result says whether the problem looks fixable and how complex the change would be.

Into the pipeline

One click converts an analyzed entry into a task. From there it follows the normal path: planning, plan approval, implementation, verification and a pull request.

Autopilot

A workspace can turn on autopilot to skip that click. Entries at or above a minimum severity, high by default, and at or below a maximum complexity, moderate by default, become tasks and start planning. At most 3 open autopilot tasks exist at a time, and a person still approves every plan.

Plan requirements

The Sentry integration needs the Team plan or above. The check happens on the server when an integration is created, so a workspace on a lower plan cannot add one through the API either.