@comfanion/usethis_search 4.4.0 → 4.5.0

package/vectorizer/usage-tracker.ts

@@ -46,6 +46,8 @@ export interface UsageData {
 }
 
 const MAX_PROVENANCE_PER_CHUNK = 20
+const MAX_TRACKED_CHUNKS = 5000      // Cap total tracked chunks to prevent unbounded growth
+const STALE_CHUNK_AGE_MS = 7 * 24 * 60 * 60 * 1000  // 7 days — evict chunks not used since
 
 // ---------------------------------------------------------------------------
 // UsageTracker
@@ -126,6 +128,40 @@ export class UsageTracker {
       stats.lastUsed = now
     }
     this.dirty = true
+
+    // Evict stale + over-cap entries periodically (every 50 searches)
+    if (this.data.totalSearches % 50 === 0) {
+      this.evictStaleChunks()
+    }
+  }
+
+  /**
+   * Evict chunks not used within STALE_CHUNK_AGE_MS, then cap at MAX_TRACKED_CHUNKS.
+   * Keeps the most recently used chunks.
+   */
+  private evictStaleChunks(): void {
+    if (!this.data) return
+    const now = Date.now()
+    const chunks = this.data.chunks
+
+    // Phase 1: remove stale (not used in 7 days)
+    for (const [id, stats] of Object.entries(chunks)) {
+      if (stats.lastUsed > 0 && now - stats.lastUsed > STALE_CHUNK_AGE_MS) {
+        delete chunks[id]
+        this.dirty = true
+      }
+    }
+
+    // Phase 2: if still over cap, evict least-used
+    const entries = Object.entries(chunks)
+    if (entries.length > MAX_TRACKED_CHUNKS) {
+      entries.sort((a, b) => a[1].lastUsed - b[1].lastUsed)
+      const toRemove = entries.length - MAX_TRACKED_CHUNKS
+      for (let i = 0; i < toRemove; i++) {
+        delete chunks[entries[i][0]]
+      }
+      this.dirty = true
+    }
   }
 
   // ---- FR-062: "where is chunk X used?" -----------------------------------

package/vectorizer.yaml

@@ -62,8 +62,8 @@ vectorizer:
     max_chunks: 100                # Max number of chunks in workspace
     attach_top_n: 10               # Top N search chunks to attach with full content
     attach_related_per_chunk: 3    # Max graph relation chunks per main chunk
-    min_score_main: 0.65           # Min score for main chunks
-    min_score_related: 0.5         # Min score for graph relation chunks
+    min_score_main: 0.5            # Min score for main chunks
+    min_score_related: 0.35        # Min score for graph relation chunks
     persist_content: false         # Save full chunk content in snapshots (debug mode)
     auto_prune_search: true        # Replace old search outputs with compact summaries
     substitute_tool_outputs: true  # Replace read() outputs when chunks in workspace