I am using AntiGravity IDE 2.0 on Win11. In my Python project the agent started a task and mentioned that it waits until the task finishes. Unfortunately that task never stopped. So this conversation does not react anymore - all my chat messages are getting queued.
I meanwhile stopped the taks which either stuck in an infinite loop or anything similar - but the conversation still is not responding. I know I can start a new one - but then the entire context of the current conversation is lost (export also does not work on this conversation).
How can I unblock/reactivate this conversation? And how can I tell within system prompt that the agent should better and periodically check its BG tasks?
Here’s the best way to rewind, along with alternative methods:
(This is written by a human)
A: Use agentapi’s send-message
This method requires your session ID. It can be found by loading the session and retrieving the conversation ID from the latest transcript directory. You can run:
Method
agentapi send-message <convid> <message: Like "wake up">
Why this may work
These are out-of-band messages. They are treated like system messages. You can use them for in-flight communication.
B: Use rewind
If you have any rewind checkpoints in this session, you can use them.
C: Reconstruct the session
This is easy: say you want to reload all data for a session, you can start a new session and invoke this:
Tell the agent to find the latest session (or do a grep) and identify it.
Ask it to load x hours or all of it.
Instruct it to batch the file and ensure it fully reads the “important context”.
This will practically restore your session.
Rationale
We used to have a bespoke save/load function to load from immediate context and native transcript system. Re-loading transcripts in full dhould help.