Codebase condensing feature doesn't work if roo accidently tried to read a very large source file #6069
Description
App Version
v0.0.6292
API Provider
VS Code Language Model API
Model Used
Claude 4 Sonnet, Claude 3.7 Sonnet, Claude 3.5 Sonnet
Roo Code Task Links (Optional)
No response
🔁 Steps to Reproduce
Setup
I am connecting to my Ubuntu linux system from my windows machine via SSH remote option. I am on the Roo Code Nightly version.
Exact Issue
Suppose if i have very large codefiles in my codebase (files with more than 5000 lines of code), roo code will try to read the file, then it will throw the error "Failed to condense " because the context got fully utilized by trying to read the file and it fully wipes out the context of task at hand. This becomes a very big pain when you are using Orchestrator mode, it delegated tasks to some Code or debug mode, and roo code will try to read a large file and gets stuck in that sub task. It's like its full memory is wiped and is really hard to make it go back to the main Orchestrator. I have explicit rules stating to not read any files above 1000 lines of code, but most of the time, it just tries to read the full file, wipes its own memory, and then it starts some random task based on the file that is open in my VS code tab. The behaviour is just wild if the "failed to condense" message happens.
💥 Outcome Summary
Expected to not wipe the Roo Code's memory or the sub task's memory completely, making it difficult to return to the orchestrator.