upstream-relation: I’m the author of Terminal Agent and I’m publish the snap.
supported-category: “HPC or orchestration agents/software for running workloads on systems without traditional users…”
reasoning: Application executes unix commands on user’s behalf in user’s workspace. By default the app allows only read only operations (ls, find, cat - with whitelisted args) in allowed workdirs; however, there are some loaded guns in case someone wants footgun themselves.
I understand that strict confinement is generally preferred over classic.
I’ve tried the existing interfaces to make the snap to work under strict confinement.
Based on the details provided, the application requires access to arbitrary system files and host-specific software. Under the Snap confinement model, these requirements fall into the unsupported category for classic confinement.
Additionally, the application does not appear to fit the specific exception category you mentioned (HPC or orchestration agents/software for running workloads on systems without traditional users where the systems are otherwise managed outside of the agent).
Please also know that classic confinement is a sensitive matter and it is reserved for mature, well-known applications published by mature, well-known entities. As of today, I believe that Terminal agent doesn’t meet this criteria.
Because the current requirements do not meet the strict criteria for classic confinement, we cannot approve this request in its current state.
I highly recommend looking into how similar tools, such as claude-code and copilot-cli, successfully utilize strict confinement.
Thank you @shishirsub10 for the response and detailed explanation.
Is there a process of asking for an exception or re-evaluation?
I agree with you that the category selected isn’t describing the application. However, of provided categories, this one in my mind fit the closest as none of classified the application well.
I have looked at the repository provided - thank you - but snaps their solve different purposes. For one, they’re more “coding agents” which are meant to work within constrained directory. Terminal agent, on the other hand, is “your helpful OS agents” helping to monitor processes and provide reports/solutions. I understand that one could “mount” a lot of directories within the snap but that somehow, I think, defeats the purpose.
I understand I’m asking for a lot. I understand that available source code is only partially helpful as there’s AI component here which might affect the actual execution. However, would there be a precedence for distribution with tag indicating that this is “a footgun” and that users should be aware of consequences?
I understand your concern and agree with the technical justification too. LLM agents may decide to use any random tool during runtime which is difficult to predict and package in advance.
And the foot-gun that you are referring to is classic confinement. And as I have already mentioned, even though the technical reason is justified for classic, classic is reserved for mature, well-known applications published by mature, well-known entries.
That is why we are unable to grant classic at this moment, sorry. We can revisit this in the future if the criteria for publisher vetting are changed.
In the meanwhile, if you have not already, I would suggest looking into existing interfaces like home (for file access), system-observe (process information) etc.
