Consider applying MIKE IO's agent-documentation work here

Placeholder — not to be actioned yet.

MIKE IO is running an exercise to make its documentation work well for AI agents writing user scripts: measure what agents actually get wrong, ship the guidance inside the wheel so it is version-matched and reachable offline, and execute every code snippet in CI so it cannot rot. The plan is [roadmap/ai-agent-documentation.md](https://github.com/DHI/mikeio/blob/main/roadmap/ai-agent-documentation.md), tracked in [DHI/mikeio#995](https://github.com/DHI/mikeio/pull/995).

modelskill has the same shape of problem — an API that agents write advisory scripts against, with idioms a model will not infer correctly on its own — so the approach should transfer. It also sits downstream of mikeio, which raises a question worth answering before copying anything: whether an agent needs one combined picture of both libraries rather than two separate guides it may only find one of.

Revisit once the mikeio work has a measured before/after. If it did not move the numbers there, it will not move them here either.

Do not act on this placeholder yet. Read roadmap/ai-agent-documentation.md and review the referenced DHI/mikeio#995 work, then wait for its measured before/after results. If the approach proves effective, determine whether modelskill needs a version-matched offline guide of its own or one combined guide covering both libraries.

Written by the indexing model from the issue text.