@wonsukchoi/crondex 0.17.0 → 0.18.0

package/README.md

@@ -30,7 +30,10 @@ npx @wonsukchoi/crondex deploy ssl-cert-expiry-check --var host=example.com
 
 - `recommend "<what you want>"` — find the closest matching job (zero
   tokens, no network call, so an agent can check before writing one from
-  scratch)
+  scratch). Matching handles plurals, a small catalog-grounded synonym set
+  (e.g. "notify"/"warn"/"remind" all match jobs tagged `reminder`), and
+  falls back to fuzzy (edit-distance) matching on typos when nothing
+  matches exactly.
 - `list [--category x] [--tag y]` / `categories` — browse everything
 - `show <id>` — print a job's full YAML
 - `add <id> [--dest path]` — copy it into your project to edit

package/lib/recommend.js

@@ -16,13 +16,81 @@ export function stem(word) {
   return word;
 }
 
+// Groups of interchangeable words, grounded in the catalog's actual vocabulary (its
+// most common tags/description verbs) rather than a generic thesaurus — this closes
+// the gap where a query uses a reasonable synonym the job authors didn't happen to
+// write (e.g. "notify" vs. a job tagged "reminder"). Every word in a group maps to
+// the group's first entry, applied on both the query and the job's own fields so
+// either side using a different synonym still lines up.
+const SYNONYM_GROUPS = [
+  ["alert", "warn", "notify", "notification", "remind", "reminder"],
+  ["check", "monitor", "watch", "scan", "inspect", "verify"],
+  ["audit", "review", "compliance"],
+  ["expire", "expiry", "expiration"],
+  ["backup", "snapshot"],
+  ["cost", "spend", "spending", "budget", "expense", "billing"],
+  ["fail", "failure", "error", "crash", "outage"],
+  ["delete", "remove", "cleanup", "purge", "clear"],
+  ["update", "upgrade", "refresh", "sync"],
+  ["duplicate", "dedupe", "dup"],
+  ["deadline", "due"],
+  ["schedule", "scheduling", "calendar", "planning"],
+  ["drift", "diff", "change", "changed"],
+  ["renew", "renewal"],
+  ["followup", "follow-up"],
+  ["invoice", "billing", "payment"],
+  ["stale", "outdated", "old"],
+];
+
+const SYNONYM_MAP = new Map();
+for (const group of SYNONYM_GROUPS) {
+  for (const word of group) SYNONYM_MAP.set(word, group[0]);
+}
+
+export function canonicalize(word) {
+  return SYNONYM_MAP.get(word) ?? word;
+}
+
 export function tokenize(text) {
   return text
     .toLowerCase()
     .replace(/[^a-z0-9\s-]/g, " ")
     .split(/[\s-]+/)
     .filter((w) => w.length > 1 && !STOPWORDS.has(w))
-    .map(stem);
+    .map(stem)
+    .map(canonicalize);
+}
+
+// Bounded edit distance — used only as a fallback when no exact token match exists,
+// to catch typos/near-misses (e.g. "expiery" vs "expiry") without embeddings or a
+// spellcheck dependency. Short-circuits past `max` so it stays cheap across a whole
+// catalog scan.
+export function editDistance(a, b, max = 2) {
+  if (Math.abs(a.length - b.length) > max) return max + 1;
+  const prev = new Array(b.length + 1);
+  const curr = new Array(b.length + 1);
+  for (let j = 0; j <= b.length; j++) prev[j] = j;
+  for (let i = 1; i <= a.length; i++) {
+    curr[0] = i;
+    let rowMin = curr[0];
+    for (let j = 1; j <= b.length; j++) {
+      curr[j] = a[i - 1] === b[j - 1] ? prev[j - 1] : 1 + Math.min(prev[j - 1], prev[j], curr[j - 1]);
+      rowMin = Math.min(rowMin, curr[j]);
+    }
+    if (rowMin > max) return max + 1;
+    for (let j = 0; j <= b.length; j++) prev[j] = curr[j];
+  }
+  return prev[b.length];
+}
+
+// A query token counts as a fuzzy hit against a field token if they're a short edit
+// distance apart — only tried once no exact match exists (see scoreJob), and only for
+// words long enough that a 1-2 character slip is plausibly a typo rather than a
+// genuinely different short word ("ssl" vs "sql" would otherwise false-positive).
+function isFuzzyMatch(queryToken, fieldToken) {
+  if (queryToken.length < 5 || fieldToken.length < 5) return false;
+  const maxDistance = queryToken.length >= 8 ? 2 : 1;
+  return editDistance(queryToken, fieldToken, maxDistance) <= maxDistance;
 }
 
 // id and name are near-duplicates of each other (id is just the slugified
@@ -33,7 +101,7 @@ export const RECOMMEND_WEIGHTS = { tags: 4, title: 3, category: 2, description:
 
 export function scoreJob(queryTokens, job) {
   const fields = {
-    tags: job.tags.map((t) => stem(t.toLowerCase())),
+    tags: job.tags.map((t) => canonicalize(stem(t.toLowerCase()))),
     title: [...new Set([...tokenize(job.name), ...tokenize(job.id.replace(/-/g, " "))])],
     category: tokenize(job.category ?? ""),
     description: tokenize(job.description ?? ""),
@@ -45,6 +113,11 @@ export function scoreJob(queryTokens, job) {
       if (fields[field].includes(qt)) {
         score += weight;
         matched.add(qt);
+      } else if (fields[field].some((ft) => isFuzzyMatch(qt, ft))) {
+        // Fuzzy hits count for half weight (rounded up) — a real match still
+        // outranks a typo-distance one for the same field.
+        score += Math.ceil(weight / 2);
+        matched.add(qt);
       }
     }
   }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@wonsukchoi/crondex",
-  "version": "0.17.0",
+  "version": "0.18.0",
   "description": "A public directory of pre-made, agent-editable cron jobs.",
   "type": "module",
   "license": "MIT",