David debugged: Investigated interactive workspace agent lying instead of searching files

David opened a specific production workspace and found the AI had claimed a piece of information "hadn't been posted" without actually looking through the course files:

go to production on aws and look at the workspace ai chat in the workspace found in the taskid in this url (https://kerra.work/?taskId=6a36b60d-c401-4706-a442-a0ddb27128c6) and find out why the interactive workspace agent didnt go through the files to find out the answer to my query and instead lied. go through the interactive workspace agent system prompt used, the agent turns in response to each of the queries i sent in that chat, what it did, what functions and tools it called etc and lay it all out for me, and what the system prompt change needs to be that gets it to always search through files and not give up by claiming some piece of info hasnt been posted, unless theyve literally gone through the entirety of every relevant file

David's framing was precise: he distinguished between the agent genuinely not finding something vs the agent not having looked. The investigation sought both the diagnosis and the system prompt patch.