Skip to content

Improve duplicate-surfacing search recall (in:title + high-recall queries) #6409

Description

@denelon

Relevant area(s)

WinGet CLI

Description of the new feature / enhancement

The agentic Duplicate Surfacing workflow (#6399 / #6400) reasons about duplicates semantically, but its only retrieval mechanism is GitHub's full-text issue search. That search de-ranks older, sparsely-worded issues, so genuine duplicates can never reach the model to be judged.

Concrete miss (found by @Trenly, #6406): on #6405 ("Support source type detection") the agent's jargon-heavy queries (source add --type PreIndexedPackage Rest autodetect, etc.) returned nothing relevant, so it no-op'd — but #1477 ("The type of a source should be automatically detected when possible") is a near-verbatim duplicate (same feature and same proposed /information-endpoint mechanism). GitHub full-text search never surfaced #1477, even for its own title words; only in:title concept searches find it.

Proposed technical implementation details

Update the workflow's "How to search" prompt guidance to raise recall:

  • Always run at least two in:title concept searches (2–3 core nouns, e.g. source type in:title).
  • Prefer short, high-recall queries; do not pile many exact CLI tokens into one query (each extra term shrinks results).
  • Vary noun/verb vocabulary (weak stemming) and scan ~20–30 results per search.

Validated via gh aw trial replaying #6405 against the live corpus: with the new guidance the agent surfaces #1477 with HIGH confidence.

Metadata

Metadata

Assignees

No one assigned

    Labels

    Issue-FeatureThis is a feature request for the Windows Package Manager client.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions