@crewhaus/tool-text 0.7.0 → 0.8.1

package/dist/lib/template.d.ts

@@ -1,7 +1,13 @@
+import type { PatternAnswer } from "./regex";
 export declare class TemplateError extends Error {
     readonly name = "TemplateError";
 }
-/** Walk a dotted path (`user.name`, `items.0.id`) through plain data. */
+/**
+ * Walk a dotted path (`user.name`, `items.0.id`) through plain data. Only
+ * the data's own fields are read: `{{constructor}}` on data with no such
+ * field is missing, where 0.7.0 rendered Object's constructor as
+ * "function Object() { [native code] }" and strict mode let it through.
+ */
 export declare function lookupPath(data: unknown, path: string): unknown;
 /**
  * Render `{{path}}` placeholders against `data`.
@@ -31,6 +37,16 @@ export type ClassifyResult = {
     readonly score: number;
     readonly matched: ReadonlyArray<string>;
     readonly scores: Readonly<Record<string, number>>;
+    /**
+     * Regex patterns with no answer (the worker's deadline, the engine giving
+     * up). When any is listed, `label` is null: a pattern that might have
+     * matched could change the winner, so neither the top label nor the
+     * default is given.
+     */
+    readonly undetermined?: ReadonlyArray<{
+        readonly pattern: string;
+        readonly reason: string;
+    }>;
 };
 /**
  * Score text against operator-written rules and return the best label.
@@ -38,5 +54,10 @@ export type ClassifyResult = {
  * Lexical only — phrase and regex matching with per-rule weights. No model
  * and no embedding, so the same message always gets the same label, which is
  * what makes it safe to route on.
+ *
+ * A regex rule's pattern is never run here: `regexAnswers` holds its answer
+ * on this text, worked out in the regex worker beforehand (the tool does
+ * that with `answerPatterns`). A pattern with no definite answer leaves the
+ * label undetermined.
  */
-export declare function classifyByRules(text: string, rules: ReadonlyArray<ClassifyRule>, threshold: number, defaultLabel: string | null): ClassifyResult;
+export declare function classifyByRules(text: string, rules: ReadonlyArray<ClassifyRule>, threshold: number, defaultLabel: string | null, regexAnswers?: ReadonlyMap<string, PatternAnswer>): ClassifyResult;

package/dist/lib/template.js

@@ -1,7 +1,12 @@
 export class TemplateError extends Error {
     name = "TemplateError";
 }
-/** Walk a dotted path (`user.name`, `items.0.id`) through plain data. */
+/**
+ * Walk a dotted path (`user.name`, `items.0.id`) through plain data. Only
+ * the data's own fields are read: `{{constructor}}` on data with no such
+ * field is missing, where 0.7.0 rendered Object's constructor as
+ * "function Object() { [native code] }" and strict mode let it through.
+ */
 export function lookupPath(data, path) {
     if (path === ".")
         return data;
@@ -16,7 +21,7 @@ export function lookupPath(data, path) {
             cur = cur[i];
             continue;
         }
-        if (typeof cur !== "object")
+        if (typeof cur !== "object" || !Object.hasOwn(cur, part))
             return undefined;
         cur = cur[part];
     }
@@ -59,18 +64,38 @@ export function renderTemplateString(template, data, strict) {
  * Lexical only — phrase and regex matching with per-rule weights. No model
  * and no embedding, so the same message always gets the same label, which is
  * what makes it safe to route on.
+ *
+ * A regex rule's pattern is never run here: `regexAnswers` holds its answer
+ * on this text, worked out in the regex worker beforehand (the tool does
+ * that with `answerPatterns`). A pattern with no definite answer leaves the
+ * label undetermined.
  */
-export function classifyByRules(text, rules, threshold, defaultLabel) {
+export function classifyByRules(text, rules, threshold, defaultLabel, regexAnswers = new Map()) {
     const haystack = text.toLowerCase();
-    const scores = {};
+    // Labels are the caller's text: a label named `constructor` or
+    // `__proto__` is a label, never an Object.prototype member.
+    const scores = Object.create(null);
     const matched = [];
+    const undetermined = [];
     for (const rule of rules) {
         const weight = rule.weight ?? 1;
         let hits = 0;
         for (const pattern of rule.patterns) {
-            const found = rule.regex === true
-                ? new RegExp(pattern, "i").test(text)
-                : haystack.includes(pattern.toLowerCase());
+            let found;
+            if (rule.regex === true) {
+                const answer = regexAnswers.get(pattern);
+                if (typeof answer !== "boolean") {
+                    undetermined.push({
+                        pattern,
+                        reason: answer?.undetermined ?? "the pattern was not answered before classifying",
+                    });
+                    continue;
+                }
+                found = answer;
+            }
+            else {
+                found = haystack.includes(pattern.toLowerCase());
+            }
             if (found) {
                 hits += 1;
                 matched.push(pattern);
@@ -79,6 +104,10 @@ export function classifyByRules(text, rules, threshold, defaultLabel) {
         if (hits > 0)
             scores[rule.label] = (scores[rule.label] ?? 0) + hits * weight;
     }
+    if (undetermined.length > 0) {
+        const best = Math.max(0, ...Object.values(scores));
+        return { label: null, score: best, matched, scores, undetermined };
+    }
     const ranked = Object.entries(scores).sort((a, b) => b[1] - a[1] || a[0].localeCompare(b[0]));
     const top = ranked[0];
     if (top === undefined || top[1] < threshold) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@crewhaus/tool-text",
-  "version": "0.7.0",
+  "version": "0.8.1",
   "type": "module",
   "description": "Deterministic text tools: regex extraction, diffing, unified-diff parsing, budget truncation, log compaction, outlining, templating, rule classification, normalization, similarity, and escaping",
   "main": "dist/index.js",
@@ -8,16 +8,18 @@
   "exports": {
     ".": {
       "types": "./dist/index.d.ts",
-      "import": "./dist/index.js"
+      "import": "./dist/index.js",
+      "default": "./dist/index.js"
     }
   },
   "scripts": {
     "test": "bun test src"
   },
   "dependencies": {
-    "@crewhaus/errors": "0.7.0",
-    "@crewhaus/tool-builder": "0.7.0",
-    "@crewhaus/tool-catalog": "0.7.0",
+    "@crewhaus/errors": "0.8.1",
+    "@crewhaus/tool-builder": "0.8.1",
+    "@crewhaus/tool-catalog": "0.8.1",
+    "@crewhaus/tool-safety": "0.8.1",
     "zod": "^3.23.8"
   },
   "devDependencies": {
@@ -41,5 +43,8 @@
   "publishConfig": {
     "access": "public"
   },
-  "files": ["dist", "README.md", "LICENSE", "NOTICE"]
+  "files": ["dist", "README.md", "LICENSE", "NOTICE"],
+  "engines": {
+    "bun": ">=1.3.14"
+  }
 }