should an MCP, OpenHands_APP, and Sandbox expect to have unique credentials and privilege levels associated with said credentials, for a unit of work? I can see arguments on either side of a binary answer at each specificity .... I don't really have a strong stance... but I'm curious what y'alls thoughts are .... I come from a universe wherein I have many different privilege levels available to me, and am expected to operate within the contextual container of least privilege required to accomplish the goal at hand... this does complicate things.... often unnecessarially... at the same time, it does also maintain a strongly scoped stupidity blast radius... WHICH I USE REGULARLY BECAUSE I MAKE A LOT OF MISTAKES... Typo is (not)literally my native language..... point being I don't think I'm perfect by any sense of the word ... and I don't expect my tools to be either ... but I do want to be mindful of just how badly they can bite me, and try to prevent future catastrophe from occurring if I can do so elegantly...

One small thing I’ve been thinking of doing, if the lack of some rarely needed logging becomes annoying, is to add a ‘trace’ level to OH logging. We could save those to files, and also purge that file at the end of the run or something.

might be cleaner to have some sorta side-endpoint thats responsible for 'consume intent / output at _ loglevel' and 'emit intent/output at_ loglevel ' .... it would be able to encrypt with keys not exposed to the app /sandbox should that be a requirement ... and be able to deny requests to said content to unauthorized requestees (say someone asking for tracelog for events unrelated to them ) but at the same time maybe this is a facet of a larger 'event / action controller ' and implementing something like this is an antipattern .... idk really