Post

LR
@lr.duangle.com

@mariozechner.at @me.snemarch.dk meanwhile i figured out the correct framing for my context compactor so the model uses it extensively: subagent launching! we pretend we're the subagent when the checkpoint is set, and when subagent_result is called, we compact, and the model buys the deception :)

0 reposts0 likes

Sune Marcher
@me.snemarch.dk

So... Do you launch a compaction every time you've invoked a subagent? Or have you replaced normal compaction with a subagent launch, and swap the current context with the subagent result?

LR
@lr.duangle.com

the model invokes the agent, the "agent" runs in the same context (we basically just switch roles for the context) until it "returns" a result, which causes the compaction. so after compaction, it looks to the context like it launched a subagent and it returned work, and it can continue with it.

LR
@lr.duangle.com

the benefit here compared to a "real" subagent invocation is that the "agent" of course sees the complete existing context so can continue exactly the way the main context would.