add-comment tool enforces rules during safe-outputs instead of during call

[strawgate]

Some tools like add-comment enforce some rules like max length, max number of mentions, etc but those rules are only checked when it's time to handle the safe outputs

They are not checked during the agent tool call. If we were able to return feedback to the agent that its comment is too long or has too many mentions in it then the Agent could make a correction.

[pelikhan]

Need to update the spec to enforce limits in the server