When a coding agent finishes, “done” leaves me with questions.
What changed? Did the project build? Which tests ran? Does the result address the original acceptance criteria?
Those questions shaped how I built Coreaven, a workflow tool for .NET projects.
1. A diff tied to the task
The first thing I want to inspect is the actual change. A summary helps me navigate, but I still need to review the code.
Coreaven connects the task, reviewed prompt, execution, and resulting GitHub PR so I can follow the work from request to result.
2. Explicit validation results
A successful agent process does not tell me whether the project builds or its tests pass.
I want to see the restore/build/test results, including failures and skipped steps. A skipped check should remain visible as skipped.
3. Evidence for individual acceptance criteria
A passing build does not prove that every requirement was implemented.
Coreaven shows acceptance-criteria evidence as Verified, Partial, or No Evidence. These states help identify what has supporting validation and what still needs attention.
They support review; they do not replace it.
4. A clear review decision
Coding runs through a Windows Runner in the user’s environment, using their own coding-agent account.
After execution, the developer can inspect the changes and validation results, review the GitHub PR, and decide whether to merge.
Try it on a real task
I’m looking for .NET developers willing to try one small task on an existing repository. I can help with setup and the first run.
What evidence do you look for before merging an AI-generated PR?
Disclosure: I’m the developer behind Coreaven. This post was drafted with AI assistance.
Top comments (0)