> ## Documentation Index
> Fetch the complete documentation index at: https://opensre.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Keep CI green with an agent

> A recurring reliability report, a bounded repair loop for one pull request, or the managed service that repairs CI for your organization.

Three ways to let OpenSRE keep working on your CI after the first report. All
of them start from the demo picker (`opensre`, or `/demo` in the shell) or from
a plain request.

## The reliability agent

Demo option **B**, "Set up an agent that improves CI/CD reliability over time",
turns the [CI reliability analysis](/docs/cicd/cicd-analytics-demo) into a recurring
check for one repository.

```text theme={null}
set up an agent that improves CI/CD reliability for Tracer-Cloud/opensre
```

* It analyzes the repository now, then schedules the report for weekdays at 08:00 local time. Later runs cover the last 7 days.
* Reports land in the shell inbox: `/loops messages`. The agent never posts to Slack or another chat channel.
* Manage it with `/loops list`, `/loops run <id>`, `/loops stop <id>` and `/loops delete <id>`.
* The loop fires while the shell is open. `/loops service install` keeps it running after you close the terminal (launchd on macOS, systemd on Linux; Windows uses `opensre cron start --service` from Task Scheduler).

## The repair loop

A repair loop watches one pull request, edits, tests and pushes fixes for its
failing GitHub Actions checks, and stops on its own:

| Rule        | Value                                                          |
| ----------- | -------------------------------------------------------------- |
| Trigger     | Every 30 seconds while active                                  |
| Attempts    | At most three failed repair attempts                           |
| Time budget | Ten minutes from scheduling                                    |
| Target      | An open pull request whose branch lives in the same repository |

Ask for it with the repository and the PR number:

```text theme={null}
schedule a CI repair loop for Tracer-Cloud/opensre pull request 6054
```

OpenSRE asks for approval before the loop starts, because it will push to the
PR branch. A pull request that is closed, or opened from a fork, is refused
before anything is scheduled, with the reason and what to choose instead.

While it runs and after it finishes:

```text theme={null}
/loops show <task-id>     the report: outcome, attempts, evidence links
/loops stop <task-id>     stop an active repair
what did the most recent CI repair run do
```

A finished run keeps its report and attempt records and removes its checkout.
Asking for the same PR again while a loop is active returns that loop with its
original deadline.

<Note>
  Want to see the loop end to end without touching a real PR? Ask for the demo repair: OpenSRE creates a small private repository named `opensre-onboarding-ci-repair-demo` with a failing check, repairs it, and closes the demo PR when the pushed checks pass. It needs a GitHub token that may create a private repository and a signed-in coding agent.
</Note>

## The managed service

Demo option **C**, "Run CI/CD repairs remotely", runs the
same repair loop on your organization's hosted gateway instead of your
machine, so it keeps running after the shell closes and uses the
organization's GitHub credential rather than yours.

What you need:

* An organization on [app.opensre.com](https://app.opensre.com) with the managed gateway provisioned from [Agent Backend settings](https://app.opensre.com/settings/agent-backend).
* The organization's GitHub token connected on the web app's Integrations page. The gateway picks up a changed token within a minute.
* Your account must be an organization admin to start work there.

Then, in the shell:

```text theme={null}
Check whether my organization's hosted gateway is running.
run the CI repair loop for Tracer-Cloud/opensre pull request 6054 on my org's hosted gateway
what did the most recent CI repair run on the gateway do
```

The gateway asks for approval through the shell before a loop starts, and
answers questions about its loops while a repair is running. See
[Deployment](/docs/guides/deployment#opensre-cloud-managed-gateway) for starting and
stopping the gateway.

## Related

* [Fix GitHub CI](/docs/cicd/github-ci-fix) — a one-off repair from the shell, no loop
* [CI/CD analytics demo](/docs/cicd/cicd-analytics-demo) — the report the reliability agent repeats
* [Shell commands](/docs/getting-started/interactive-shell-commands) — everything `/loops` can do