Agents are now responsible for creating Triggers from PaperRemotes (replacing can_edit semantics)
All @ai semantics are handled by the agent
PaperRemote can veto an edit before it starts, allowing for debouncing (replacing/refactoring get_next_edit_request)
PaperRemote perform_edit now does one thing, namely writing the edit that the agent gave
Main items left to do for mvp demo:
still need to generalize Triggers
consider specifying a Protocol for the lifecycle of a macro... and figure out how to hook my contexts into guidance
need to generalize edits (like Triggers) to be more than just str outputs that edit the same lines as the trigger
everything in PaperRemote is using default_doc. better handling different docs within a paper will go hand in hand with updating the trigger/action data structures
Main changes:
@ai
semantics are handled by the agentMain items left to do for mvp demo:
context
s into guidance