Hi,
I’ve read the other threads, and appreciate the suggestions about “hook-like” workarounds.
There are some mechanisms that real hooks would support. Here’s one example, in case it motivates the antigravity developers to prioritize hooks as a feature
Context-mode uses a combination of MCP and hooks interfaces to track what an agent is doing, reduce the context the agent has to manage, and intercept certain requests to provide needed historical context as-needed, when-needed.
I realize individual problem cases can be handled with skills and rules, but it feels like context-mode would prevent a lot of re-discovery effort (and mistakes) by the agent in the general case.