I’m Öcal, the developer behind Coreaven. I built it to keep the steps around an AI coding task connected: requirements, repository context, prompt preparation, execution, validation results, and pull request review.
This post introduces my own product and the engineering choices behind its initial .NET workflow.
Start with a task you can review
The workflow starts with a task and optional acceptance criteria. Coreaven prepares a repository-aware prompt that you can inspect and edit before running your coding agent.
For an existing .NET codebase, the relevant context includes more than a file name: symbols, their relationships, repository instructions, and the exact revision being changed all matter. Coreaven uses a Roslyn-based code index to help select that context.
Keep execution separate from the working checkout
The coding agent runs through a Windows Runner in your environment, using your own coding-agent account. Each coding job uses an isolated Git worktree.
That gives the task a specific base commit and a separate workspace for its changes. You can follow execution activity and inspect the resulting files without treating “the agent finished” as proof that the task succeeded.
Make validation visible
For the .NET workflow, validation follows restore → build → test. The results remain associated with the execution so you can review them alongside the changes.
A passing build and test run is useful evidence, but it does not establish that every requirement is met. Acceptance criteria and the actual diff still need review. Coreaven keeps validation, review, and GitHub delivery visible as separate stages.
Review the result before merging
The complete path is:
Task → repository-aware prompt → coding execution → validation → GitHub PR → human review
You review the prompt before execution and the changes before merging. The goal is to make it easier to trace a pull request back to the task that produced it and the validation that ran.
What I’m working on next
The current onboarding includes connecting GitHub, installing the Runner, and authenticating the coding agent. That setup is a real part of the experience, and reducing its friction is one of my priorities.
I’m looking for .NET developers and small teams willing to try one small task on an existing repository. I’m happy to help with setup and the first run.
You can explore Coreaven at https://coreaven.com.
What information do you want alongside an AI-generated PR to make the review useful? I’d especially like feedback on repository context, validation results, and the handoff to human review.
