If you ask the AI user to review code in comment, it will never approve or request for changes for the PR. You will need to add the AI user as a reviewer instead.

I did. The first AI response (scout) is after I created the PR with the AI agent as reviewer

Maybe this screenshot provides more info:

The system prompt already tells the AI model to call the PR approve tool upon accepted review. It works well for models with good tool call capabilities, such as Claude Haiku etc.

Maybe another place to force the LLM to call a taskComplete with PR action before it completes its review.

I was wondering if there is any place in the folders, or if a configuration page could be added, that allows users to see/tweak the system prompts used by OneDev?

Local models used by users are most likely not as powerful as the cloud models (I can run max 24B while cloud models go up to 256B+), and in my limited experience, some need some "extra coaching" to get something done.

Another possibility would be to have an overview of available calls, so that I can tweak the AI User's system prompt.

Maybe another place to force the LLM to call a taskComplete

In my previous post here I had an issue with chat text disappearing. My workaround was to forbid calling "taskComplete", could this be what is now causing the PR to not go through?

You may AI settings of the AI user, and there is a system prompt setting. "taskComplete" tool is not currently not being used for code review tasks.

State changed as build OD-7256 (15.0.0) is successful