Triage a Sentry issue

Connect Sentry first. A workflow step is an ordinary session, so it reaches Sentry the same way a chat does.

Build the chain

From Workflows → New workflow → Blank workflow, add one input, one context document, and three steps.

Input — issue, required, described as "the Sentry issue URL or short ID".

Context document — slug triage, written by the first step:

# Triage
 
## What broke
 
## Where it comes from
 
## Proposed fix

Step 1, Agent — "Read the issue"

Pull @input:issue from Sentry: the stack trace, the release it started in, how
often it fires, and which users it hits. Write what broke and where it comes
from into @doc:triage, with the file and line the trace points at.

Step 2, Agent — "Find the cause"

Read the code @doc:triage points at and work out why it fails. Update the
document with the actual cause and the smallest fix you would make. Do not
change any code yet.

Step 3, Human in the loop — "Decide"

Read @doc:triage. Approve if the diagnosis and the proposed fix are right, or
redo the previous step with what it got wrong.

Run it

Press Run, paste the issue URL, and pick the repository the service lives in. Leave the placement on New worktree.

Read triage under Documents when the run stops on your step.