We need to implement an auto-context memory system with compression capabilities for managing conversation…
We need to implement an auto-context memory system with compression capabilities for managing conversation…: a task in LegoFlow-SWE (Harbor dataset). Problem symptoms: - Conversations exceed token limits silently, causing errors in downstream model calls. - No mechanism to keep recent context…
The task
**Problem symptoms:** - Conversations exceed token limits silently, causing errors in downstream model calls. - No mechanism to keep recent context while summarizing older messages. - Tool interactions (consecutive tool use/result blocks) bloat the memory without meaningful compression. - Original messages are lost…
Part of Lego-X/LegoFlow-SWE.