Define a self-healing pipeline agent

Let's build as much context as possible to make self-healing as automated as possible. Start with a failed run and its logs. Then add asset descriptions, quality checks, ownership, lineage, and any unit tests you have. Each piece gives the agent more information to find the cause and run an allowed recovery.

Its configuration determines which recovery actions the agent can take automatically. A cautious setup lets it investigate and prepare pull requests. A more automated setup can recover a known issue, such as late source data, without requiring approval for every retry.

What to do

• Name the first pipeline the agent may investigate.
• Name the pipeline owner. If code changes need review, name the reviewer too.
• Choose whether the agent may suggest a repair, open a pull request, or run a named recovery step automatically.
• Put these rules in the project AGENTS.md before connecting the agent to production alerts.

Why this matters: the same failure can come from a source delay, bad data, or a code bug. The agent needs clear limits before it chooses an action.