akm-cli 0.9.0-beta.30 → 0.9.0-beta.31

package/CHANGELOG.md

@@ -6,6 +6,19 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/).
 
 ## [Unreleased]
 
+## [0.9.0-beta.31] — 2026-06-20
+
+### Changed
+
+- **#632 — recombine now filters junk tags structurally.** Frontmatter tags that
+  are pure numbers, dates (`20260529`), short hex hashes (`002c624c`), version
+  strings (`0.8.0`, `v2`), single chars, or common English stopwords (`is`, `the`,
+  `for`, `when`, …) carry no topical signal and never form a recombine cluster.
+  Unlike `excludeTags` (a fixed project list), this catches the OPEN-ENDED junk —
+  every new date or commit hash — with no config upkeep. Exposed as `isJunkTag`.
+  On the live stash this turns the recombine cluster set from generic 66–171-member
+  buckets into tight topical clusters (`auth`, `architecture`, `patterns`, …).
+
 ## [0.9.0-beta.30] — 2026-06-20
 
 ### Changed / Fixed

package/dist/commands/improve/recombine.js

@@ -67,6 +67,69 @@ const DEFAULT_CONFIRM_THRESHOLD = 2;
  */
 const DEFAULT_RECOMBINE_OVERLAP = 0.7;
 // ── Clustering by relatedness (NOT similarity) ────────────────────────────────
+/**
+ * #632 — English stopwords that occasionally leak into frontmatter tags
+ * (`is`, `the`, `for`, …). They carry no topical signal, so a cluster keyed on
+ * one is meaningless. Lowercased; matched case-insensitively.
+ */
+const JUNK_STOPWORD_TAGS = new Set([
+    "a",
+    "an",
+    "and",
+    "the",
+    "to",
+    "of",
+    "in",
+    "on",
+    "for",
+    "is",
+    "are",
+    "be",
+    "no",
+    "not",
+    "or",
+    "if",
+    "it",
+    "as",
+    "at",
+    "by",
+    "we",
+    "us",
+    "do",
+    "so",
+    "when",
+    "then",
+    "than",
+    "with",
+    "from",
+    "this",
+    "that",
+    "uses",
+    "use",
+    "via",
+]);
+/**
+ * #632 — a tag carries no clustering signal (and must be skipped) when it is
+ * purely a number / date / hash / version string, a single char, or a common
+ * stopword. Unlike `excludeTags` (a fixed project list), this catches the
+ * OPEN-ENDED junk — every new date or commit hash — without config upkeep.
+ */
+export function isJunkTag(tag) {
+    const t = tag.trim().toLowerCase();
+    if (t.length <= 1)
+        return true;
+    if (JUNK_STOPWORD_TAGS.has(t))
+        return true;
+    if (/^\d+$/.test(t))
+        return true; // pure numbers + dates: 2026, 05, 23, 20260529
+    if (/^v?\d+(?:\.\d+)+$/.test(t))
+        return true; // versions: 0.8.0, v1.2
+    if (/^v\d+$/.test(t))
+        return true; // v0, v2
+    if (/^[0-9a-f]{4,}$/.test(t) && /\d/.test(t))
+        return true; // short hex hashes: 002c624c, 192d
+    return false;
+}
 /**
  * Build relatedness clusters from the memory pool. Clustering is driven purely
  * by shared tags / graph entities — it MUST NOT use embedding similarity, so
@@ -114,6 +177,8 @@ export function buildRelatednessClusters(entries, opts) {
             for (const tag of entry.entry.tags ?? []) {
                 if (excludeTags.has(tag))
                     continue;
+                if (isJunkTag(tag))
+                    continue; // #632 — skip numeric/date/hash/version/stopword junk
                 add(`tag:${tag}`, entry);
             }
         }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "akm-cli",
-  "version": "0.9.0-beta.30",
+  "version": "0.9.0-beta.31",
   "type": "module",
   "description": "akm (Agent Knowledge Management) — A package manager for AI agent skills, commands, tools, and knowledge. Works with Claude Code, OpenCode, Cursor, and any AI coding assistant.",
   "keywords": [