The AI narration layer isn't reachable from either frontend

[Muawiya-contact]

diskern-core has an ai feature with a trait, an error type, and a
working TemplateNarrator:

pub trait AiProvider: Send + Sync {
    fn narrate(&self, findings: &[Finding]) -> Result<String, AiError>;
}

Nothing enables the feature, no frontend calls it, and no provider is
implemented — so the narration layer the README describes as
"AI narrates" does not exist in a build anyone can run.

The contract is already written and is the part worth keeping: the
provider gets findings the engine has already decided on and returns
prose. It cannot change a verdict, a score, or trigger an action, and
the app has to work identically with no provider configured. Any PR here
has to preserve that — narration is a view over the report, not a step
in the pipeline.

Roughly in order of usefulness:

1. Wire TemplateNarrator into the CLI behind --explain. No network,
   no key, no new dependency — and it proves the seam works before
   anyone builds a provider on it.
2. An OllamaProvider posting to http://localhost:11434. Local, free,
   nothing leaves the machine.
3. BYO-key providers, which need somewhere safe to keep a key and a
   settings surface in the app — bigger, and worth its own issue.

Start at 1 if you want the smallest useful piece.

crates/diskern-core/src/ai.rs · crates/diskern-core/Cargo.toml:13

[Muawiya-contact]

Taking this one myself — it needs more context on the internals than an issue description can reasonably hand over, so I'd rather not leave someone stuck on it.

Not off-limits for discussion though. If you have ideas or want to talk through the approach, comment here.

If you're after something to pick up, the issues labelled level: beginner are scoped to a single file with the reasoning already worked out.