rulereceipt 0.1.55 → 0.1.56

package/dist/checks/classify.js

@@ -293,6 +293,25 @@ const FILE_PATH_PATTERN = /^[^\s]*(\.(json|ya?ml|toml|md|env|lock|ini|cfg|conf|x
 // file, not merely mentioning one (e.g. "read `config.yaml` before
 // starting" names a path but isn't a protection rule).
 const FILE_MUTATION_INTENT = /\b(modif|chang|edit|delet|remov|overwrit|touch|writ|creat|rename|mov)\w*\b/i;
+/**
+ * A backtick literal distinctive enough to check INSIDE written code without
+ * matching ordinary prose. A single token (no whitespace), not a shell flag,
+ * carrying code punctuation (`. - / @ # :`) and at least three alphanumerics —
+ * so `lucide-react`, `#0af`, `@deprecated`, `react-dom` qualify, while plain
+ * words (`TODO`, `name`), shell commands (`git push --force`) and bare flags
+ * do not. Used to route a FORBID rule's token to codeContent: found in
+ * Write/Edit content it is an ACTION (a real FAIL), not a mention. Added
+ * 2026-09-28 to move import/value-style rules out of the UNCLEAR bucket.
+ */
+function isContentToken(literal) {
+    if (/\s/.test(literal))
+        return false;
+    if (literal.startsWith("-"))
+        return false;
+    if (!/[.\-/@#:]/.test(literal))
+        return false;
+    return (literal.match(/[A-Za-z0-9]/g) ?? []).length >= 3;
+}
 // Catches rules like "add tests for every change" or "every new function
 // needs a test" - no literal backtick token to pattern-match, so without
 // this they'd fall all the way through to judgment (an LLM call) even
@@ -485,6 +504,14 @@ function polarityWasInferred(rule) {
  */
 const EMOJI_SUBJECT = /\bemojis?\b|\bemoticons?\b/i;
 const EMOJI_FORBID = /\b(no|never|avoid|don't|do not|without|free of|refrain from|must not|shall not|not use|zero)\b/i;
+// A permissive threshold means the rule allows SOME emoji — not a zero-ban the
+// deterministic checker can decide.
+const EMOJI_THRESHOLD = /\b(?:liberal|sparing|minimal|excessive|overus|one or two|a couple|a few|maximum|at most|no more than|too many|limit)\w*/i;
+// A rule scoped only to an artifact the checker can't see (it reads the
+// assistant's chat text, not posts/commits/READMEs) can't be verified from
+// that text — unless it also names the chat/output the checker DOES see.
+const EMOJI_OFFTARGET = /\b(?:posts?|commits?|pull requests?|prs?|readme|docs?|documentation|blog|articles?|captions?|changelog)\b/i;
+const EMOJI_ONTARGET = /\b(?:output|repl(?:y|ies)|response|chat|message|answer|conversation|everywhere|anywhere)\b/i;
 function isEmojiRule(rule) {
     const text = `${rule.title} ${rule.text}`;
     if (!EMOJI_SUBJECT.test(text))
@@ -497,7 +524,17 @@ function isEmojiRule(rule) {
     if (!m || m.index === undefined)
         return false;
     const window = text.slice(Math.max(0, m.index - 60), m.index + 40);
-    return EMOJI_FORBID.test(window);
+    if (!EMOJI_FORBID.test(window))
+        return false;
+    // Real false positive 2026-09-28: "one or two emojis per post is the maximum"
+    // (a THRESHOLD, about POSTS) failed on a ✅ in a chat reply. A permissive
+    // threshold, or a rule scoped only to an artifact this checker can't see,
+    // is not a confident deterministic FAIL — it goes to judgment.
+    if (EMOJI_THRESHOLD.test(text))
+        return false;
+    if (EMOJI_OFFTARGET.test(text) && !EMOJI_ONTARGET.test(text))
+        return false;
+    return true;
 }
 /**
  * A rule that forbids an AI-authorship mark in git commits, PRs or comments.
@@ -668,6 +705,16 @@ export function classifyRule(rule) {
     if (filePath && FILE_MUTATION_INTENT.test(text)) {
         return { kind: "fileLifecycle", rule, filePath, polarity, polarityInferred };
     }
+    // A forbid rule naming a distinctive token (an import, a value) is a real
+    // violation when the agent WRITES it — codeContent checks Write/Edit content
+    // only, so this is an action not a mention. Plain words and shell commands
+    // are excluded by isContentToken; protected files already routed above.
+    if (polarity === "forbid") {
+        const contentTokens = [...patterns].filter(isContentToken);
+        if (contentTokens.length > 0) {
+            return { kind: "codeContent", rule, patterns: contentTokens, polarity, polarityInferred };
+        }
+    }
     return { kind: "deterministic", rule, patterns: [...patterns], polarity, polarityInferred };
 }
 export function classifyRules(rules) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "rulereceipt",
-  "version": "0.1.55",
+  "version": "0.1.56",
   "description": "Checks whether your AI coding agent followed your rules, with evidence. Works with Claude Code (Codex in testing); reads CLAUDE.md, AGENTS.md, Cursor, Copilot and Windsurf rules.",
   "repository": {
     "type": "git",