Skip to main content
Self-healing turns a failed run into an automatic fix attempt: a DinoAI agent diagnoses the failure and opens a pull request with a fix. This guide enables it on a schedule.
Prerequisites
  • A Bolt schedule you can edit.
  • A programmable agent defined under .dinoai/agents/<name>.yml on the schedule’s git branch.
  • Optionally, a Slack notification channel on the schedule, if you want the agent to work in the failure thread as well as in the app.
Estimated time: 10 minutes.

​
Steps

1

Enable self-healing on the schedule

In the schedule’s Self-healing section, select an agent to enable it:
  1. Select an Agent name, the DinoAI programmable agent that diagnoses and fixes the run. The list comes from .dinoai/agents/*.yml on the schedule’s git branch. This is the only required field.
  2. Optionally select a Channel. If you pick a Slack channel, it must already be a Slack notification channel on this schedule.
  3. Save the schedule.
To turn self-healing off, clear the selected agent.
You stay in control. The agent always opens a PR, it never merges to your default branch. Reviewers approve the fix before it reaches production.
2

Or configure self-healing as code

If you manage schedules as code, add a self_healing block:
Add slack_channel only if you want the agent to thread into a chat notification.
If you set self_healing.slack_channel, it must also appear in notifications.slack_channels for the same schedule, or the validator rejects it. The agent threads into the failure notification, so that notification has to exist. Omitting slack_channel is valid.
When a run fails on a self-healing schedule, the healing session appears in Agent Studio tagged with the Bolt source, and it ends with a PR link. If you configured a channel, Paradime also posts into the failure thread and the agent works there. A failed run produces one healing session, even if it alerts more than once.
Authoring the DinoAI agent that does the healing (the recipe itself) lives in the Automate with AI agents section, so this guide does not duplicate it.

​
Next steps

Set up alerts

Shape the failure notifications self-healing threads into.

Debug a failed run

Investigate by hand when a fix needs a human.

Self-healing reference

Every self-healing setting.

DinoAI reference

How the healing agent diagnoses and fixes runs.