Ruben Fiszel ad53673a28 fix(ai-chat): improve resource-type search tool description and scoring (#10272) ...

The global-mode search_resource_types tool described its query as a
substring match, but the backend does semantic embedding search. Rewrite
the description so the LLM sends natural-language intent instead of literal
name guesses.

Apply the top-score trim to query_resource_types so a strong match isn't
diluted by weakly-related types that merely clear the 0.75 similarity floor.
The trim was previously inlined only in query_hub_scripts; extract it into a
shared trim_to_top_score helper, use it from both search paths, and add a
unit test pinning the 5% cutoff boundary.

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

2026-07-23 00:54:18 +02:00

..

src

fix(ai-chat): improve resource-type search tool description and scoring (#10272)

2026-07-23 00:54:18 +02:00

Cargo.toml

refactor: extract windmill-api into subcrates for parallel compilation (#7845)

2026-02-07 22:12:55 +00:00