agent-context.json

{
  "version": 1,
  "summary": "Improves command-palette matching by normalizing query text and ranking stronger matches ahead of loose substring hits.",
  "files": [
    {
      "path": "src/normalize.ts",
      "summary": "Centralizes query cleanup before any matching happens.",
      "annotations": [
        {
          "newRange": [1, 3],
          "summary": "Adds one normalization helper for whitespace, case, and dashed shortcut terms.",
          "rationale": "This lets the search layer reason about one normalized token shape instead of repeating slightly different cleanup logic in multiple places.",
          "author": "sonnet"
        }
      ]
    },
    {
      "path": "src/search.ts",
      "summary": "Scores and sorts matches instead of returning the first loose substring list.",
      "annotations": [
        {
          "newRange": [15, 35],
          "summary": "Prefix and exact keyword matches now outrank weaker substring hits before the result list is sorted.",
          "rationale": "The old behavior made every match look equally good, which was fine for filtering but weak for command-palette ranking where the top result should usually be the most obvious intent.",
          "author": "sonnet"
        },
        {
          "newRange": [20, 27],
          "summary": "Worth checking the score floor — could mask edge cases.",
          "rationale": "The scoring thresholds (4, 3, 2, 1) look good but validate that zero-score items are properly filtered out.",
          "author": "prism"
        }
      ]
    },
    {
      "path": "src/index.ts",
      "summary": "Keeps the preview intentionally short.",
      "annotations": [
        {
          "newRange": [1, 8],
          "summary": "The preview now shows only the top three ranked commands.",
          "rationale": "Once ranking is reliable, the preview can stay compact and let the best results carry the review without flooding the UI.",
          "author": "prism"
        }
      ]
    },
    {
      "path": "test/search.demo.ts",
      "summary": "Locks in normalized query handling and result ordering.",
      "annotations": [
        {
          "newRange": [1, 8],
          "summary": "The test covers a dashed query form so the new normalization helper has a visible behavioral contract.",
          "rationale": "Without a test that exercises `short-cuts` specifically, it would be easy to regress the helper and still pass on simpler substring-only cases.",
          "author": "sonnet"
        }
      ]
    }
  ]
}