An incident for engineering: context in the channel, the fix with a person

What the agent gathers

• The Sentry issue, via the Sentry connection in the MCP catalog: title, culprit, and the first frames of the stack.
• Git log on the repo for that day, so the brief names the commits that touched the tax path.
• Open Jira issues that mention checkout or tax, so on-call does not rediscover a ticket from the afternoon.

It does not SSH into production. It does not roll back. It does not merge. Those actions are not on this agent.

The message in #oncall

The post is a brief. What broke, the Sentry link, the two commits from the afternoon, the Jira key filed at 16:10, and a suggested first read: the tax change, not the whole checkout service. The incident-response skill supplies the shape of that note. The skill does not grant Git push.

If a pull request is already open, and Bitbucket is connected, the agent can add the link. Commenting on the pull request waits until someone on call asks for it.

The next morning's review is a different job

Code review is the quieter version of the same pattern. A pull request opens, the agent reads the diff, and it comments on tests, risky patterns, and missing docs. A senior engineer still approves the merge. The agent review is a first pass, not a gate you hide behind.

Which tools an engineering agent should have at all is in The tools your agents can actually call.