The issue I've had with gates like this is that the agent has to choose to do it, and often it will forget to. If you move your checks inside hooks rather than putting them in agent prompts, you can configure it so agents literally can't do certain things unless they pass a check. Alternatively, the hook can inject things into the prompt at the ideal time, which increases the probability that the agent will conform to it.
The issue I've had with gates like this is that the agent has to choose to do it, and often it will forget to. If you move your checks inside hooks rather than putting them in agent prompts, you can configure it so agents literally can't do certain things unless they pass a check. Alternatively, the hook can inject things into the prompt at the ideal time, which increases the probability that the agent will conform to it.
[flagged]
[flagged]