I'd like a way to, in isolation, test plumbing functionality .... as in manually operate MCPs, test access to api's etc ... "Perform operation _ as if I were _ agent ..." sorta things .... also, I wonder .... do y'all suppose there's merit to the idea of thinking about OH work as: goal( plan( procedure/task( subtask ( step () ) ) ) ) and then making it possible to expose debug output intuitively for any scope wherein an action encountered any error.... but if no error occurred, expect to purge the debuglog... as in don't EMIT logs at debug level to (logging system whatever) but securely retain them briefly for the purpose of diagnostics, then purge? For example... when my sandbox VM was trying to hit my gitlab instance via an MCP it encountered a 401 unauthorized error from my gitlab rig... I don't want to keep a packetcapture.... but I'd like to be able to interrogate what call was made to see why it errored... it doesn't feel logical to try and get that level of logging from (remote endpoints) because our collective access to serverside telemetry's gonna be spotty at best .... trying to do so on the component-side would result in high fragmentation of effort ..... ergo someplace within openhands feels like the logical place to keep track of the 'I wanted to do this ... but encountered an error and pivoted .... which worked, but I retained the output in high fidelity for optional review..... would you like to .... [ review ] [ purge ]