"""Prompts held outside the code that uses them, as TOML.

One file per processor, named `<processor_name>.toml`. Each prompt sits at
`[prompts.<key>]` with its body in a `text` field, and `load_prompt(processor,
key)` returns that body stripped. Adding a prompt means adding a key to the
TOML, not editing Python.

This covers the table-inference processor behind `quber analyze` and `quber
batch`, and nothing else. The agents under `quber.agents` — the grid locator,
the structure vetter, the status inspector, and the rest of the Set-of-Mark and
figure paths — define their system prompts inline at the call site instead.
Look there first when changing how one of those behaves.
"""

from quber.prompts.loader import load_prompt

__all__ = ["load_prompt"]