The agent gets stuck frequently and doesn't often correct itself on errors. We should make the agent loop more robust, be giving clear error reports to the agent and only returning an error to the user when the agent can not recover. I'm open to fundamentally changing the current loop/harness.
The agent gets stuck frequently and doesn't often correct itself on errors. We should make the agent loop more robust, be giving clear error reports to the agent and only returning an error to the user when the agent can not recover. I'm open to fundamentally changing the current loop/harness.