my agent is writing code and when I load a file into memory and it changes, I want to remove the old version from the context in order to prevent confusion in the future. (only one version of the code is visible to the agent and that prevents mistakes, but requires deleting messages from the messages list). Also to maximize cached token usage (I know deleting message will ruin the cache, but this is for a different use case). To do that I need to make sure that different agents get the same list of messages. If one call conditionally injects a message that goes out to the LLM, the other agent will need to have duplicate logic to inject the same message in the same spot, or the cache is ruined. Right now I'm doing the injection outside of BAML code for this use case.

The collector API will get you this! https://docs.boundaryml.com/guide/baml-advanced/collector-track-tokens