The platform should support a DAG of jobs triggered by the API run. It can include agentic analysis, actions, and deterministic steps. It can run on the same or multiple machines.
One example is the CI repair workflow:
[Prepare data -> Analyze the failure] -> Notify sheriffs in Slack -> [Generate a fix -> Build locally] -> Run TRY push and wait -> Revise the fix if failed and repeat -> Notify the developer
This one looks linear for now, but we might want to parallelize things later on to reduce latency. Like Run Try push and notify the developer that it's running, and maybe continue some deep review at the same time. This definitely adds quite a bit of complexity, so we can enforce the process to be linear until we run into issues.
There are a few things that might benefit from this:
#6579
#6621
#6623
#6663
The platform should support a DAG of jobs triggered by the API run. It can include agentic analysis, actions, and deterministic steps. It can run on the same or multiple machines.
One example is the CI repair workflow:
[Prepare data -> Analyze the failure] -> Notify sheriffs in Slack -> [Generate a fix -> Build locally] -> Run TRY push and wait -> Revise the fix if failed and repeat -> Notify the developer
This one looks linear for now, but we might want to parallelize things later on to reduce latency. Like Run Try push and notify the developer that it's running, and maybe continue some deep review at the same time. This definitely adds quite a bit of complexity, so we can enforce the process to be linear until we run into issues.
There are a few things that might benefit from this:
#6579
#6621
#6623
#6663