I don't think the metric 'any code that decreases agility as the project matures' is quite the same thing. I think most code lowers agility, as it is increasing complexity and the number of interactions and entanglements across your codebase. Adding features decreases agility. These are not 'technical debt'.

I always enjoy Ward Cunningham's discussion of this: http://c2.com/cgi/wiki?WardExplainsDebtMetaphor (Ward first used the metaphor)

Specifically, he is NOT talking about 'dirty code' or just badly written code, that's taking shortcuts and I believe it is never a good idea in software. We're talking about building good code but doing so before we fully understand the problem, often with the aim of learning more about the problem, especially from user feedback.

What I take Ward's discussion to mean is that "technical debt" is accrued when writing code for a problem you do not YET completely understand (e.g. almost all 'first pass' code). The debt occurs when your code's existing understanding is static (because it is in source control), and your teams understanding grows. The difference is your debt. You can continue building on the code's not-quite-right understanding for the moment which is carrying the debt forward and paying the costs of fighting a not-quite-right model, but at some point it becomes worth while to re-factor based on what your team has learned about your problem, and thus pay off the debt.