Executing AI generated commands automatically is dangerous. Of course that will probably seem like a quaint notion someday not too long from now.
But it could be useful for pal to act as a stage in a text pipeline, and in that case, it's no problem to limit access for those commands.
- Use
/pipe to indicate that the first command suggestion returned should be executed, with stdin wired up
- Run the command suggestion in a container (user will need to enable this and specify docker/podman/?)
- These should probably be deterministic, which would imply caching the results to make sure the same one is always run
command1 | pal /pipe transform this data | command2
Executing AI generated commands automatically is dangerous. Of course that will probably seem like a quaint notion someday not too long from now.
But it could be useful for
palto act as a stage in a text pipeline, and in that case, it's no problem to limit access for those commands./pipeto indicate that the first command suggestion returned should be executed, withstdinwired up