I'd actually argue that this is the wrong way around. Being able to git diff legislation is very good, but mainly for humans, who need tools to make work tractable.
An agent will use whatever interface is available, including no interference, won't give up due to the task being too onerous, and can be scaled by paying for more tokens. It could simply brute force diff the raw text, writing bespoke Python and regexes, etc.
LLMs make brute forcing intellect possible. It's humans who need finely crafted tools.