Sentry with DevFlow: setup, scope and limits

DevFlow polls one Sentry project for unresolved issues, records each new issue as an attention item and has an agent rate its severity. Critical and high items get a deeper read-only analysis against your code. Setup is 5 steps in Workspace Settings → Integrations, on the Team plan and above.

Kind
Issue triage
Where to set it up
Workspace Settings → Integrations
Plan
Team and above
Agent roles
sentry-triage, attention-analysis
Verified
1 Oct 2026

Setup steps

  1. Create an auth token. In Sentry, create an auth token that can read the project's issues. DevFlow sends it as a Bearer token to /api/0/projects/<org>/<project>/issues/; the code does not list token scopes, it only needs that request to return 200.
  2. Open the Integrations tab. Go to Workspace Settings → Integrations. Below the Team plan, the tab shows an upgrade notice instead of the integration list; its "View plans" button switches to the Billing tab.
  3. Fill in the connection. Click "Add Integration", choose Sentry and fill "Organization Slug", "Project Slug" and "Auth Token". For a self-hosted server, also set "Base URL (for self-hosted)", which otherwise defaults to https://sentry.io, then click "Add Integration".
  4. Test the connection. Click "Test" on the new card. "Connected!" means the issues request succeeded with your token; otherwise the card shows the status code and error text Sentry returned.
  5. Work the attention items. New issues arrive as attention items after the next poll. Open one to read its triage, click "Analyze" for an item that was not analyzed automatically, then "Convert to Task" or "Dismiss".

What it does

Every 2 minutes the poller requests the project's issue list with query=is:unresolved&sort=date and skips issues last seen before the previous poll. Each issue is stored once, keyed by its issue ID, so a later poll never creates a duplicate.

A new attention item goes to the sentry-triage role, which has 2 minutes to return a severity (critical, high, medium or low), a one-paragraph summary and the affected area of the code. Items rated critical or high then go to attention-analysis: a 10-minute pass with only Read, Grep and Glob that reports root cause, fixability, fix complexity and affected files.

Items move through new, triaging, triaged, analyzing and analyzed, and end as converted or dismissed. The card in the Integrations tab shows when the last sync ran, and its buttons force a sync, run the test, pause or resume polling, edit the settings or delete the connection.

Converting an item creates an idle task from the issue and its analysis. Planning starts the usual way, so a person still approves the plan before code is written.

Requirements

Limits

FAQ

Does DevFlow fix Sentry errors without a human?

No. A Sentry item becomes a task only when someone converts it or when the opt-in autopilot does, and in both cases a person approves the plan and marks the task done.

Which model triages Sentry issues in DevFlow?

In DevFlow, the workspace's own configuration decides. The sentry-triage and attention-analysis roles can each be given a model in the per-agent model overrides under Workspace Settings → Execution.

What does the DevFlow autopilot do with Sentry items?

When enabled, the DevFlow autopilot converts analyzed Sentry items that are fixable and within the workspace's severity and complexity limits, then starts planning. By default it keeps at most 3 autopilot tasks open.

Sources

Every fact above comes from these files in the DevFlow repository.