Proposal #3:
Hacked bots remember what they kill and Tux can replay their "memories" by
talking to them and running a program that Tux wrote, gaining experience from
new kills (since the program was last run). Maybe you would also need to run the
program on the bot before it starts recording memories.

The program could be created when you get novice at programming (from Sorenson)
and hack a certain number of bots?

It could evolve with combinations of the rank in programming, number of bots
hacked, experience gained through the program, and level of "Reverse
Engineering" (mentioned in proposal #2).
The amount of experience you gain could be determining by what gets recorded in
the current version (image, sound, gyroscope/movement sensors, commands from the
A.I., or any other input that a bot might provide).
Inspired by proposal #1, when the program is advanced enough it could send
experience directly to Tux in real time through botnet (and by consequence only
when the bot is not too damaged).

As the downside, you lose unseen experience when the bot dies and lose the time
needed to run the program.
Inspired by proposal #2, you could get paralyzed or slowed for some time when
you gain experience from the replays (because you are busy processing them).