Skip to main content
This guide sets up an automation that runs on each new Sentry issue. The agent rates the issue, finds the cause in your code, posts a short summary to Slack, and opens a pull request with the fix. False alarms are not posted.

​
Before you start

You need:
  • Sentry connected, with the projects you want to watch.
  • GitHub connected, with the GitHub App granted access to the repositories behind those projects.
  • Slack connected, if you want the summary in a channel.
Connect them under Integrations, or from the automation editor when the template asks for them.

​
Create the automation

1

Pick the template

Open Automations. Under Start from a template, filter by Bug triage and click Alert on critical Sentry issues. The editor opens with the trigger, instructions, and the Sentry, GitHub, and Slack connectors filled in. See Templates.
2

Choose the Sentry projects

The template adds a Sentry New issue trigger. Under Project, pick the projects to watch, up to 100.
3

Exclude environments

Under Environment, clear the environments that should not start a run, such as staging or development. The default is All environments. Environments that appear in Sentry later are included.
4

Add repositories

Under Repositories, click Add repository and pick the code behind the projects. The first one is the agent’s working directory.
5

Add a Slack channel

Under Notifications, click Add notification, then Post to Slack, and pick the channel for the summaries.
6

Choose a model and save

Click Choose model in the instructions toolbar, pick a model, then click Save. See Models and harnesses.

​
Trigger options

The Sentry trigger has three events: To narrow what starts a run, select fewer projects or exclude more environments. See Triggers.

​
Adjust the instructions

The template tells the agent to:
  1. Rate the issue from SEV-1 to SEV-3, and stay quiet on a complete false positive.
  2. Post a short message with the severity, what broke, the suspected commit and author, and who is affected.
  3. Add the root cause and proposed fix in the thread, open a pull request, and link it there.
Edit the instructions under Agent instructions to match your team. For example, change the severity scale, say which errors are known noise, or tell the agent to skip the pull request for some services.

​
What a run does

Each new issue starts one run, titled with the Sentry short ID and the issue title. The run gets a fresh sandbox with your repositories checked out. The agent receives the issue details and can read more events and traces through the Sentry connector. The agent can post to the notification channel during the run. If it does not, its final message is posted there with a View run link. When the agent decides an issue is not worth reporting, it skips the notification, and the run page shows the reason. A failed run is always reported. When the agent opens a pull request, it appears as a card in the run and on the PRs page. See Runs and Pull requests.

​
Review the pull request

Pull requests open without an approval step. Use your branch protection and review rules as the gate before merge. To ask for changes, submit a review on GitHub. The review continues the run that opened the pull request. The agent checks each comment, pushes fixes to the same branch, and replies to each comment. You can also send a follow-up from the run page. See Follow-ups from GitHub reviews.

​
Test it

Click Test in the automation header and paste a Sentry issue link or the error details. Your message is treated as the trigger event, so you can check the agent’s output before real issues arrive.