@comfanion/usethis_search 3.0.0-dev.23 → 3.0.0-dev.24

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@comfanion/usethis_search",
-  "version": "3.0.0-dev.23",
+  "version": "3.0.0-dev.24",
   "description": "OpenCode plugin: semantic search with graph-based context (v3: graph relations, 1-hop context, LSP + regex analyzers)",
   "type": "module",
   "main": "./index.ts",

package/tools/search.ts

@@ -1,8 +1,8 @@
 /**
- * Semantic Code Search Tool (v2)
+ * Semantic Code Search Tool (v3)
  *
  * Uses local embeddings + LanceDB vector store via bundled vectorizer.
- * v2: hybrid search, metadata filtering, rich result metadata.
+ * v3: simplified agent API — 5 params, config-driven defaults, smart filter.
  * Index data is stored in `.opencode/vectors/<index>/`.
  */
 
@@ -10,7 +10,82 @@ import { tool } from "@opencode-ai/plugin"
 import path from "path"
 import fs from "fs/promises"
 
-import { CodebaseIndexer } from "../vectorizer/index.ts"
+import { CodebaseIndexer, getSearchConfig } from "../vectorizer/index.ts"
+
+// ── Extension → language mapping (for filter parsing) ─────────────────────
+const EXT_TO_LANG: Record<string, string> = {
+  go: "go", py: "python", ts: "typescript", tsx: "typescript",
+  js: "javascript", jsx: "javascript", mjs: "javascript", cjs: "javascript",
+  rs: "rust", java: "java", kt: "kotlin", swift: "swift",
+  c: "c", cpp: "cpp", h: "c", hpp: "cpp", cs: "csharp",
+  rb: "ruby", php: "php", scala: "scala", clj: "clojure",
+  md: "markdown", mdx: "markdown", txt: "text",
+  yaml: "yaml", yml: "yaml", json: "json", toml: "toml",
+}
+const LANG_NAMES = new Set(Object.values(EXT_TO_LANG))
+
+/**
+ * Parse the `filter` param into path prefix and/or language filter.
+ *
+ * Supported formats:
+ *   "internal/domain/"    → pathPrefix = "internal/domain"
+ *   "*.go"                → language = "go"
+ *   ".go"                 → language = "go"
+ *   "go"                  → language = "go"
+ *   "internal/*.go"       → pathPrefix = "internal", language = "go"
+ *   "internal/**\/*.go"   → pathPrefix = "internal", language = "go"
+ *   "service"             → pathContains = "service"
+ */
+function parseFilter(filter: string): {
+  pathPrefix?: string
+  language?: string
+  pathContains?: string
+} {
+  if (!filter) return {}
+
+  const f = filter.trim()
+
+  // "internal/**/*.go" or "internal/*.go" → path + extension
+  const globMatch = f.match(/^([^*]+?)(?:\/\*\*)?\/?\*\.(\w+)$/)
+  if (globMatch) {
+    const prefix = globMatch[1].replace(/\/+$/, "")
+    const ext = globMatch[2]
+    return {
+      pathPrefix: prefix,
+      language: EXT_TO_LANG[ext] || undefined,
+    }
+  }
+
+  // "*.go" or ".go" → extension only
+  const extMatch = f.match(/^\*?\.(\w+)$/)
+  if (extMatch) {
+    const ext = extMatch[1]
+    return { language: EXT_TO_LANG[ext] || undefined }
+  }
+
+  // "go", "python", "typescript" → language name
+  const lower = f.toLowerCase()
+  if (LANG_NAMES.has(lower)) {
+    return { language: lower }
+  }
+  // "go" could also be ext
+  if (EXT_TO_LANG[lower]) {
+    return { language: EXT_TO_LANG[lower] }
+  }
+
+  // Ends with "/" → path prefix
+  if (f.endsWith("/")) {
+    return { pathPrefix: f.replace(/\/+$/, "") }
+  }
+
+  // Contains "/" → path prefix (e.g. "internal/domain")
+  if (f.includes("/")) {
+    return { pathPrefix: f.replace(/\/+$/, "") }
+  }
+
+  // Anything else → substring match on file path
+  return { pathContains: f }
+}
 
 export default tool({
   description: `Search the codebase semantically. Use this to find relevant code snippets, functions, or files based on meaning, not just text matching.
@@ -22,84 +97,51 @@ Available indexes:
 - searchAll: true - Search across all indexes
 
 Examples:
-- "authentication logic" → finds auth-related code
-- "database connection handling" → finds DB setup code
-- "how to deploy" with index: "docs" → finds deployment docs
-- "API keys" with index: "config" → finds config with API settings
-- search({ query: "tenant", path: "internal/domain/" }) → searches only in internal/domain/`,
+- search({ query: "authentication logic" })
+- search({ query: "how to deploy", index: "docs" })
+- search({ query: "tenant management", filter: "internal/domain/" })
+- search({ query: "event handling", filter: "*.go" })
+- search({ query: "API routes", filter: "internal/**/*.go" })
+- search({ query: "metrics", searchAll: true })`,
 
   args: {
     query: tool.schema.string().describe("Semantic search query describing what you're looking for"),
-    index: tool.schema.string().optional().default("code").describe("Index to search: code, docs, config, or custom name"),
-    limit: tool.schema.number().optional().default(10).describe("Number of results to return (default: 10)"),
+    index: tool.schema.string().optional().default("code").describe("Index to search: code, docs, config"),
+    limit: tool.schema.number().optional().describe("Number of results (default from config, typically 10)"),
     searchAll: tool.schema.boolean().optional().default(false).describe("Search all indexes instead of just one"),
-    freshen: tool.schema.boolean().optional().default(true).describe("Auto-update stale files before searching (default: true)"),
-    includeArchived: tool.schema.boolean().optional().default(false).describe("Include archived files in results (default: false). Files are archived if in /archive/ folder or have 'archived: true' in frontmatter."),
-    // v2 params
-    hybrid: tool.schema.boolean().optional().describe("Enable hybrid search (vector + BM25 keyword matching). Improves exact keyword recall."),
-    fileType: tool.schema.string().optional().describe("Filter by file type: 'code', 'docs', or 'config'"),
-    language: tool.schema.string().optional().describe("Filter by language: 'typescript', 'python', 'markdown', etc."),
-    modifiedAfter: tool.schema.string().optional().describe("Filter: only files modified after this ISO date (e.g. '2024-01-01')"),
-    modifiedBefore: tool.schema.string().optional().describe("Filter: only files modified before this ISO date"),
-    tags: tool.schema.string().optional().describe("Filter by frontmatter tags (comma-separated, e.g. 'auth,security')"),
-    minScore: tool.schema.number().optional().default(0.35).describe("Minimum relevance score (0-1). Results below this threshold are dropped. Default: 0.35"),
-    path: tool.schema.string().optional().describe("Filter by file path prefix (e.g. 'internal/domain/', 'src/components'). Only returns files under this path."),
+    filter: tool.schema.string().optional().describe("Filter results by path or language. Examples: 'internal/domain/', '*.go', 'internal/**/*.go', 'service'"),
   },
 
   async execute(args) {
     const projectRoot = process.cwd()
 
     try {
-      let allResults: any[] = []
-      const limit = args.limit || 10
+      // Load config defaults (parsed from vectorizer.yaml)
+      const cfg = getSearchConfig()
+      const limit = args.limit || cfg.default_limit || 10
       const indexName = args.index || "code"
+      const minScore = cfg.min_score ?? 0.35
+      const includeArchived = cfg.include_archived ?? false
+
+      // Parse filter into path/language constraints
+      const filterParsed = args.filter ? parseFilter(args.filter) : {}
 
-      // Build search options from v2 params
+      // Build search options — hybrid is always from per-index config
       const searchOptions: Record<string, any> = {}
-      if (args.hybrid != null) searchOptions.hybrid = args.hybrid
-
-      // Normalize fileType: support extensions (*.go, .go) and language names (go, python)
-      // fileType field stores "code" | "docs" | "config", so map user-friendly values
-      if (args.fileType) {
-        const ft = args.fileType.replace(/^\*?\.?/, "").toLowerCase()
-        const extToLanguage: Record<string, string> = {
-          go: "go", py: "python", ts: "typescript", tsx: "typescript",
-          js: "javascript", jsx: "javascript", mjs: "javascript", cjs: "javascript",
-          rs: "rust", java: "java", kt: "kotlin", swift: "swift",
-          c: "c", cpp: "cpp", h: "c", hpp: "cpp", cs: "csharp",
-          rb: "ruby", php: "php", scala: "scala", clj: "clojure",
-          md: "markdown", mdx: "markdown", txt: "text",
-          yaml: "yaml", yml: "yaml", json: "json", toml: "toml",
-        }
-        // Also accept full language names
-        const langNames = new Set([
-          "go", "python", "typescript", "javascript", "rust", "java", "kotlin",
-          "swift", "c", "cpp", "csharp", "ruby", "php", "scala", "clojure",
-          "markdown", "text", "yaml", "json", "toml",
-        ])
-
-        if (ft === "code" || ft === "docs" || ft === "config") {
-          searchOptions.fileType = ft
-        } else if (extToLanguage[ft]) {
-          searchOptions.language = extToLanguage[ft]
-        } else if (langNames.has(ft)) {
-          searchOptions.language = ft
-        } else {
-          searchOptions.fileType = ft // pass through as-is
+      if (filterParsed.language) searchOptions.language = filterParsed.language
+
+      // Freshen from config (default: false — auto_index handles it)
+      if (cfg.freshen) {
+        try {
+          const tempIndexer = await new CodebaseIndexer(projectRoot, indexName).init()
+          await tempIndexer.freshen()
+          await tempIndexer.unloadModel()
+        } catch {
+          // non-fatal — search can proceed without freshen
         }
       }
 
-      if (args.language) searchOptions.language = args.language
-      if (args.modifiedAfter) searchOptions.modifiedAfter = args.modifiedAfter
-      if (args.modifiedBefore) searchOptions.modifiedBefore = args.modifiedBefore
-      if (args.tags) searchOptions.tags = args.tags.split(",").map((t: string) => t.trim()).filter(Boolean)
-
-      // Auto-freshen stale files before searching
-      if (args.freshen !== false) {
-        const tempIndexer = await new CodebaseIndexer(projectRoot, indexName).init()
-        await tempIndexer.freshen()
-        await tempIndexer.unloadModel()
-      }
+      let allResults: any[] = []
 
       if (args.searchAll) {
         const tempIndexer = await new CodebaseIndexer(projectRoot, "code").init()
@@ -107,21 +149,17 @@ Examples:
         await tempIndexer.unloadModel()
 
         if (indexes.length === 0) {
-          return `No indexes found. Create one with: codeindex({ action: "reindex", index: "code" })`
+          return `No indexes found. The codebase needs to be indexed first.\n\nRun the CLI: bunx usethis_search reindex`
         }
 
         for (const idx of indexes) {
           const indexer = await new CodebaseIndexer(projectRoot, idx).init()
-          if (args.freshen !== false) {
-            await indexer.freshen()
-          }
-          const results = await indexer.search(args.query, limit, args.includeArchived, searchOptions)
+          const results = await indexer.search(args.query, limit, includeArchived, searchOptions)
           allResults.push(...results.map((r: any) => ({ ...r, _index: idx })))
           await indexer.unloadModel()
         }
 
         allResults.sort((a, b) => {
-          // Prefer combinedScore (hybrid), fall back to L2→similarity conversion
           const scoreA = a._combinedScore ?? (a._distance != null ? Math.max(0, 1 - a._distance / 2) : 0)
           const scoreB = b._combinedScore ?? (b._distance != null ? Math.max(0, 1 - b._distance / 2) : 0)
           return scoreB - scoreA
@@ -145,40 +183,40 @@ Examples:
         }
 
         const indexer = await new CodebaseIndexer(projectRoot, indexName).init()
-        const results = await indexer.search(args.query, limit, args.includeArchived, searchOptions)
+        const results = await indexer.search(args.query, limit, includeArchived, searchOptions)
         allResults = results.map((r: any) => ({ ...r, _index: indexName }))
         await indexer.unloadModel()
       }
 
       // ── Score cutoff — drop low-relevance results ──────────────────────────
-      const minScore = args.minScore ?? 0.35
       allResults = allResults.filter(r => {
         const score = r._combinedScore ?? (r._distance != null ? Math.max(0, 1 - r._distance / 2) : 0)
         return score >= minScore
       })
 
-      // ── Path filter — restrict to subtree ──────────────────────────────────
-      if (args.path) {
-        const prefix = args.path.replace(/\/+$/, "") // normalize trailing slash
+      // ── Filter — apply path/language constraints from `filter` param ───────
+      if (filterParsed.pathPrefix) {
+        const prefix = filterParsed.pathPrefix
         allResults = allResults.filter(r => r.file && r.file.startsWith(prefix))
       }
+      if (filterParsed.pathContains) {
+        const needle = filterParsed.pathContains.toLowerCase()
+        allResults = allResults.filter(r => r.file && r.file.toLowerCase().includes(needle))
+      }
+      // Language filter is already passed to searchOptions above, but double-check
+      // in case vectorizer didn't filter (e.g. docs index has no language field)
+      if (filterParsed.language) {
+        allResults = allResults.filter(r => !r.language || r.language === filterParsed.language || r.language === "unknown")
+      }
 
       // ── Reranking — boost results where query keywords appear in text ──────
-      // Also store score components for breakdown display
       const queryKeywords = args.query.toLowerCase().split(/\s+/).filter((w: string) => w.length > 2)
       for (const r of allResults) {
-        // Vector score (L2 → similarity)
         const vectorScore = r._distance != null ? Math.max(0, 1 - r._distance / 2) : 0
         r._vectorScore = vectorScore
-
-        // BM25 component (present only in hybrid mode — embedded in _combinedScore)
-        // If _combinedScore exists and differs from vectorScore, the difference is BM25 contribution
         r._bm25Component = r._combinedScore != null ? Math.max(0, r._combinedScore - vectorScore) : 0
-
-        // Base score before keyword boost
         const baseScore = r._combinedScore ?? vectorScore
 
-        // Keyword matching
         const text = (r.content || "").toLowerCase()
         const matchedKeywords: string[] = []
         if (queryKeywords.length > 0) {
@@ -208,25 +246,24 @@ Examples:
         }
       }
 
-      // Sort groups by best chunk score, take top N unique files
       const sortedGroups = [...fileGroups.values()]
         .sort((a, b) => (b.best._finalScore ?? 0) - (a.best._finalScore ?? 0))
         .slice(0, limit)
 
       if (sortedGroups.length === 0) {
         const scope = args.searchAll ? "any index" : `index "${indexName}"`
-        return `No results found in ${scope} for: "${args.query}" (min score: ${minScore})\n\nTry:\n- Different keywords\n- Lower minScore threshold: search({ query: "...", minScore: 0.2 })\n- Enable hybrid search: search({ query: "...", hybrid: true })`
+        const filterNote = args.filter ? ` with filter "${args.filter}"` : ""
+        return `No results found in ${scope}${filterNote} for: "${args.query}" (min score: ${minScore})\n\nTry:\n- Different keywords or phrasing\n- Remove or broaden the filter\n- search({ query: "...", searchAll: true })`
       }
 
       // ── Confidence signal ──────────────────────────────────────────────────
       const topScore = sortedGroups[0].best._finalScore ?? 0
       const scope = args.searchAll ? "all indexes" : `index "${indexName}"`
-      const hybridLabel = args.hybrid ? " [hybrid]" : ""
-      const pathLabel = args.path ? ` path:"${args.path}"` : ""
-      let output = `## Search Results for: "${args.query}" (${scope}${hybridLabel}${pathLabel})\n\n`
+      const filterLabel = args.filter ? ` filter:"${args.filter}"` : ""
+      let output = `## Search Results for: "${args.query}" (${scope}${filterLabel})\n\n`
 
       if (topScore < 0.45) {
-        output += `> **Low confidence results.** Best score: ${topScore.toFixed(3)}. These results may not be relevant to your query.\n> Try more specific keywords, different phrasing, or hybrid: true.\n\n`
+        output += `> **Low confidence results.** Best score: ${topScore.toFixed(3)}. These results may not be relevant to your query.\n> Try more specific keywords or different phrasing.\n\n`
       }
 
       for (let i = 0; i < sortedGroups.length; i++) {
@@ -235,7 +272,7 @@ Examples:
         const indexLabel = args.searchAll ? ` [${r._index}]` : ""
         const chunkNote = chunks.length > 1 ? ` (${chunks.length} matching sections)` : ""
 
-        // v2: show rich metadata when available
+        // Rich metadata
         const metaParts: string[] = []
         if (r.language && r.language !== "unknown") metaParts.push(r.language)
         if (r.heading_context) metaParts.push(`"${r.heading_context}"`)
@@ -243,7 +280,7 @@ Examples:
         if (r.class_name) metaParts.push(`class: ${r.class_name}`)
         const metaLine = metaParts.length > 0 ? ` (${metaParts.join(", ")})` : ""
 
-        // Score breakdown: vector + bm25 + keywords
+        // Score breakdown
         const breakdownParts: string[] = [`vec: ${(r._vectorScore ?? 0).toFixed(2)}`]
         if (r._bm25Component > 0.005) breakdownParts.push(`bm25: +${r._bm25Component.toFixed(2)}`)
         if (r._keywordBonus > 0.005) breakdownParts.push(`kw: +${r._keywordBonus.toFixed(2)}`)
@@ -261,7 +298,7 @@ Examples:
         output += content
         output += "\n```\n"
 
-        // Show second-best chunk from same file if available (brief)
+        // Second-best chunk hint
         if (chunks.length > 1) {
           const second = chunks.find((c: any) => c !== r)
           if (second) {

package/vectorizer/index.ts

@@ -85,6 +85,15 @@ let HYBRID_CONFIG = { ...DEFAULT_HYBRID_CONFIG };
 let METRICS_ENABLED = false;
 let CACHE_ENABLED = true;
 
+// ── Search defaults (exposed to tool layer) ──────────────────────────────────
+const DEFAULT_SEARCH_CONFIG = {
+  freshen: false,           // Don't freshen on every search — auto_index handles it
+  min_score: 0.35,          // Minimum relevance score cutoff
+  include_archived: false,  // Exclude archived files by default
+  default_limit: 10,        // Default result count
+};
+let SEARCH_CONFIG = { ...DEFAULT_SEARCH_CONFIG };
+
 // ── Graph config (v3) ───────────────────────────────────────────────────────
 const DEFAULT_GRAPH_CONFIG = {
   enabled: true,
@@ -135,6 +144,10 @@ function defaultVectorizerYaml() {
     `  search:\n` +
     `    hybrid: true\n` +
     `    bm25_weight: 0.3\n` +
+    `    freshen: false              # Don't re-index on every search (auto_index handles it)\n` +
+    `    min_score: 0.35             # Minimum relevance score cutoff\n` +
+    `    include_archived: false     # Exclude archived files\n` +
+    `    default_limit: 10           # Default number of results\n` +
     `\n` +
     `  # Graph-based context (v3)\n` +
     `  graph:\n` +
@@ -299,6 +312,11 @@ async function loadConfig(projectRoot) {
         enabled: parseBool(ss, "hybrid", false),
         bm25_weight: parseNumber(ss, "bm25_weight", 0.3),
       };
+      // Extended search defaults
+      SEARCH_CONFIG.freshen = parseBool(ss, "freshen", DEFAULT_SEARCH_CONFIG.freshen);
+      SEARCH_CONFIG.min_score = parseNumber(ss, "min_score", DEFAULT_SEARCH_CONFIG.min_score);
+      SEARCH_CONFIG.include_archived = parseBool(ss, "include_archived", DEFAULT_SEARCH_CONFIG.include_archived);
+      SEARCH_CONFIG.default_limit = parseNumber(ss, "default_limit", DEFAULT_SEARCH_CONFIG.default_limit);
     }
 
     // ── Parse quality config ────────────────────────────────────────────────
@@ -1305,4 +1323,8 @@ function getEmbeddingModel() {
   return EMBEDDING_MODEL;
 }
 
-export { CodebaseIndexer, INDEX_PRESETS, getEmbeddingModel };
+function getSearchConfig() {
+  return SEARCH_CONFIG;
+}
+
+export { CodebaseIndexer, INDEX_PRESETS, getEmbeddingModel, getSearchConfig };