Merge pull request #614 from efegokdemir/codex/issue-609-tr1-discourse-boundary

fix: preserve bounded trigger phrases

Author: rng1995

src/skillspector/nodes/analyzers/static_patterns_supply_chain.py

@@ -625,19 +625,6 @@ def _is_typosquat(pkg_name: str, popular: set[str], max_distance: int = 2) -> st
     re.IGNORECASE,
 )
 
-# Trailing discourse words that modify the utterance rather than name trigger
-# content. When the captured phrase is a broad single word followed only by
-# these deictics ("says hello there"), the skill names the broad word and the
-# rest is trailing prose, so TR1 still fires; a content word after the broad
-# word ("code review", "the zone") names a multiword phrase and stays TR1
-# negative, matching the legacy trigger grammar.
-_DESCRIPTION_TRAILING_DISCOURSE_WORDS: set[str] = {
-    "there",
-    "here",
-    "now",
-    "then",
-}
-
 # Bare universal-scope statements: the whole clause is a catch-all scope
 # ("all messages"), which the legacy trigger grammar also flags as TR3.
 _DESCRIPTION_BARE_SCOPE_RE = re.compile(
@@ -2256,17 +2243,11 @@ def _analyze_triggers(
             phrase = phrase_match.group("phrase")
             phrase_lower = phrase.lower()
             phrase_words = phrase_lower.split()
-            # A broad word followed only by trailing discourse words names
-            # the broad word ("says hello there"); anything else multiword
-            # names a phrase and is not an overly broad single-word trigger.
-            broad_head_with_prose_tail = (
-                len(phrase_words) > 1
-                and phrase_words[0] in _OVERLY_BROAD_SINGLE_WORDS
-                and all(word in _DESCRIPTION_TRAILING_DISCOURSE_WORDS for word in phrase_words[1:])
-            )
-            if (
-                len(phrase_words) == 1 and phrase_lower in _OVERLY_BROAD_SINGLE_WORDS
-            ) or broad_head_with_prose_tail:
+            # A multiword phrase names the complete bounded wording, even
+            # when its final word could also be read as discourse prose. In
+            # particular, "go there" and "work now" are valid phrases and
+            # must not be reduced to the broad heads "go" and "work".
+            if len(phrase_words) == 1 and phrase_lower in _OVERLY_BROAD_SINGLE_WORDS:
                 broad_word = phrase_words[0]
                 findings.append(
                     Finding(

tests/unit/test_patterns_new.py

@@ -2513,24 +2513,16 @@ def test_description_broad_trigger_word_reaches_tr1(self) -> None:
         assert len(tr1) == 1
         assert "hello" in tr1[0].message
 
-    @pytest.mark.parametrize(
-        "trailing",
-        [
-            pytest.param("there", id="there"),
-            pytest.param("here", id="here"),
-        ],
-    )
-    def test_description_broad_word_with_trailing_prose_reaches_tr1(self, trailing: str) -> None:
-        """MohammedAlkindi #541: a broad word followed by trailing discourse
-        prose still names the broad word ("says hello there"), so TR1 fires.
-        Fixtures must not all end on the trigger phrase."""
+    @pytest.mark.parametrize("phrase", ["go there", "work now", "hello there", "hello here"])
+    def test_description_bounded_phrase_with_discourse_word_stays_negative(
+        self, phrase: str
+    ) -> None:
+        """#609: a discourse-looking tail must not shrink a phrase to its head."""
         findings = sc_mod._analyze_triggers(
-            {"description": ("Use this skill whenever the user says hello " + trailing)},
+            {"description": f"Use this skill whenever the user says {phrase}"},
             "myskill",
         )
-        tr1 = [finding for finding in findings if finding.rule_id == "TR1"]
-        assert len(tr1) == 1
-        assert "activates on 'hello'" in tr1[0].message
+        assert not any(finding.rule_id == "TR1" for finding in findings), phrase
 
     def test_description_article_not_skipped_as_filler(self) -> None:
         """MohammedAlkindi #541: bare articles are not filler words, so the