Merge branch 'master' into fix/ipinfo-deprecation-skip-none

Author: rennf93

.gitignore

@@ -209,4 +209,5 @@ __marimo__/
 .DS_Store
 manual_testing/
 ZZZ/
-.claude/
+.claude/
+tests/attack_simulation/out/

Makefile

@@ -231,6 +231,11 @@ local-test:
 	@find . | grep -E "(__pycache__|\\.pyc|\\.pyo|\\.pytest_cache|\\.ruff_cache|\\.mypy_cache)" | xargs rm -rf
 
 
+.PHONY: attack-sim
+attack-sim:
+	@uv run python -m tests.attack_simulation
+
+
 .PHONY: integration-test
 integration-test:
 	@INTEGRATION_TESTS=1 REDIS_URL=$${REDIS_URL:-redis://localhost:6379} IPINFO_TOKEN=$${IPINFO_TOKEN:-test_token} REDIS_PREFIX=$${REDIS_PREFIX:-test:guard_core:} uv run pytest tests/integration -m integration -v

guard_core/detection_engine/preprocessor.py

@@ -50,13 +50,9 @@ def __init__(
     )
     _HEX_ESCAPE_RE = re.compile(r"\\x([0-9a-fA-F]{2})")
     _UNICODE_ESCAPE_RE = re.compile(r"\\u([0-9a-fA-F]{4})")
-    _SQL_BLOCK_COMMENT_INNER_UPPER_RE = re.compile(
-        r"(?<=[A-Z])/\*.*?\*/(?=[A-Z])", re.DOTALL
+    _SQL_BLOCK_COMMENT_STRIP_RE = re.compile(
+        r"(?<!\w)/\*(?!!).*?\*/|/\*(?!!).*?\*/(?!\w)", re.DOTALL
     )
-    _SQL_BLOCK_COMMENT_INNER_LOWER_RE = re.compile(
-        r"(?<=[a-z])/\*.*?\*/(?=[a-z])", re.DOTALL
-    )
-    _SQL_BLOCK_COMMENT_RE = re.compile(r"/\*.*?\*/", re.DOTALL)
     _SQL_LINE_COMMENT_RE = re.compile(r"(--|#)[^\n]*")
 
     async def _send_preprocessor_event(
@@ -122,6 +118,9 @@ def normalize_unicode(self, content: str) -> str:
             "\u2216": "\\",
             "\uff1c": "<",
             "\uff1e": ">",
+            "\uff1b": ";",
+            "\uff5c": "|",
+            "\uff06": "&",
         }
 
         for char, replacement in lookalikes.items():
@@ -277,9 +276,7 @@ def _replace(match: re.Match[str]) -> str:
         return self._UNICODE_ESCAPE_RE.sub(_replace, content)
 
     def _strip_sql_comments(self, content: str) -> str:
-        content = self._SQL_BLOCK_COMMENT_INNER_UPPER_RE.sub("", content)
-        content = self._SQL_BLOCK_COMMENT_INNER_LOWER_RE.sub("", content)
-        content = self._SQL_BLOCK_COMMENT_RE.sub(" ", content)
+        content = self._SQL_BLOCK_COMMENT_STRIP_RE.sub(" ", content)
         content = self._SQL_LINE_COMMENT_RE.sub(" ", content)
         return content
 

guard_core/handlers/suspatterns_handler.py

@@ -26,6 +26,8 @@
 _CTX_SENSITIVE_FILE = frozenset({"url_path", "request_body", "unknown"})
 _CTX_CMS_PROBING = frozenset({"url_path", "request_body", "unknown"})
 _CTX_RECON = frozenset({"url_path", "unknown"})
+_CTX_PROTO_POLLUTION = frozenset({"query_param", "request_body", "unknown"})
+_CTX_CODE_INJECTION = frozenset({"query_param", "request_body", "unknown"})
 _CTX_ALL = frozenset({"query_param", "header", "url_path", "request_body", "unknown"})
 
 
@@ -47,6 +49,8 @@
         "sensitive_file",
         "cms_probing",
         "recon",
+        "proto_pollution",
+        "code_injection",
     }
 )
 
@@ -67,6 +71,8 @@
     "sensitive_file": _CTX_SENSITIVE_FILE,
     "cms_probing": _CTX_CMS_PROBING,
     "recon": _CTX_RECON,
+    "proto_pollution": _CTX_PROTO_POLLUTION,
+    "code_injection": _CTX_CODE_INJECTION,
 }
 
 
@@ -122,6 +128,19 @@ class SusPatternsManager:
             _CTX_SQLI,
             "sqli",
         ),
+        (r"\w/\*(?!!)[^*]*\*/\w", _CTX_SQLI, "sqli"),
+        (r"(?i)(?:OR|AND)\s+'[\w\d]*'='[\w\d]*'?", _CTX_SQLI, "sqli"),
+        (
+            r"(?i)(?:^|;)\s*(?:DROP|TRUNCATE|ALTER)\s+(?:TABLE|DATABASE|SCHEMA)\b",
+            _CTX_SQLI,
+            "sqli",
+        ),
+        (
+            r"(?i);\s*(?:INSERT\s+INTO|UPDATE\s+\w+\s+SET|DELETE\s+FROM)\b",
+            _CTX_SQLI,
+            "sqli",
+        ),
+        (r"(?i)\bORDER\s+BY\s+\d+\b", _CTX_SQLI, "sqli"),
         (r"(?:\.\.\/|\.\.\\)(?:\.\.\/|\.\.\\)+", _CTX_DIR_TRAVERSAL, "dir_traversal"),
         (
             r"(?:/etc/(?:passwd|shadow|group|hosts|motd|issue|mysql/my.cnf|ssh/"
@@ -162,6 +181,16 @@ class SusPatternsManager:
             _CTX_CMD_INJECTION,
             "cmd_injection",
         ),
+        (
+            r"[;|&]\s*(?:ls|cat|rm|id|whoami|uname|wget|curl|nc|netcat|socat|bash|sh|python|perl)\b",
+            _CTX_CMD_INJECTION,
+            "cmd_injection",
+        ),
+        (
+            r"(?i)\b(?:nc|netcat|ncat)\s+-[a-z]*e\b|/dev/tcp/\d",
+            _CTX_CMD_INJECTION,
+            "cmd_injection",
+        ),
         (
             r"(?:php|data|zip|rar|file|glob|expect|input|phpinfo|zlib|phar|ssh2|"
             r"rar|ogg|expect)://[^\s]+",
@@ -194,6 +223,11 @@ class SusPatternsManager:
             "nosql",
         ),
         (r"(?:\{\s*\$[a-zA-Z]+\s*:\s*(?:\{|\[))", _CTX_NOSQL, "nosql"),
+        (
+            r'"\$(?:where|gt|gte|lt|lte|ne|eq|regex|in|nin|all|size|exists|type|mod|options|expr|jsonSchema)"\s*:',
+            _CTX_NOSQL,
+            "nosql",
+        ),
         (
             r"(?i)filename=[\"'].*?\.(?:php\d*|phar|phtml|exe|jsp|asp|aspx|sh|"
             r"bash|rb|py|pl|cgi|com|bat|cmd|vbs|vbe|js|ws|wsf|msi|hta)[\"\']",
@@ -216,6 +250,12 @@ class SusPatternsManager:
             _CTX_TEMPLATE,
             "template",
         ),
+        (r"<%[=#]?[^%]*%>", _CTX_TEMPLATE, "template"),
+        (
+            r"\$\{[^}]*(?:@[\w.]+@|\b\w+\s*\(|\d+\s*[*/%+\-]\s*\d+)[^}]*\}",
+            _CTX_TEMPLATE,
+            "template",
+        ),
         (
             r"[\r\n]\s*(?:HTTP\/[0-9.]+|Location:|Set-Cookie:)",
             _CTX_HTTP_SPLIT,
@@ -334,6 +374,16 @@ class SusPatternsManager:
         (r"(?:^|/)autodiscover/", _CTX_RECON, "recon"),
         (r"^/dns-query(?:\?|$)", _CTX_RECON, "recon"),
         (r"(?:^|/)\.git/(?:refs|index|HEAD|objects|logs)(?:/|$)", _CTX_RECON, "recon"),
+        (
+            r"(?:__proto__|constructor)\s*(?:\[\s*[\"']prototype[\"']\s*\]|\.\s*prototype)|[\"']__proto__[\"']\s*:",
+            _CTX_PROTO_POLLUTION,
+            "proto_pollution",
+        ),
+        (
+            r"System\.Diagnostics\.Process\.Start\s*\(|System\.Reflection\.|Assembly\.Load\s*\(",
+            _CTX_CODE_INJECTION,
+            "code_injection",
+        ),
     ]
 
     patterns: list[str] = [p[0] for p in _pattern_definitions]

guard_core/sync/detection_engine/preprocessor.py

@@ -50,13 +50,9 @@ def __init__(
     )
     _HEX_ESCAPE_RE = re.compile(r"\\x([0-9a-fA-F]{2})")
     _UNICODE_ESCAPE_RE = re.compile(r"\\u([0-9a-fA-F]{4})")
-    _SQL_BLOCK_COMMENT_INNER_UPPER_RE = re.compile(
-        r"(?<=[A-Z])/\*.*?\*/(?=[A-Z])", re.DOTALL
+    _SQL_BLOCK_COMMENT_STRIP_RE = re.compile(
+        r"(?<!\w)/\*(?!!).*?\*/|/\*(?!!).*?\*/(?!\w)", re.DOTALL
     )
-    _SQL_BLOCK_COMMENT_INNER_LOWER_RE = re.compile(
-        r"(?<=[a-z])/\*.*?\*/(?=[a-z])", re.DOTALL
-    )
-    _SQL_BLOCK_COMMENT_RE = re.compile(r"/\*.*?\*/", re.DOTALL)
     _SQL_LINE_COMMENT_RE = re.compile(r"(--|#)[^\n]*")
 
     def _send_preprocessor_event(
@@ -122,6 +118,9 @@ def normalize_unicode(self, content: str) -> str:
             "\u2216": "\\",
             "\uff1c": "<",
             "\uff1e": ">",
+            "\uff1b": ";",
+            "\uff5c": "|",
+            "\uff06": "&",
         }
 
         for char, replacement in lookalikes.items():
@@ -277,9 +276,7 @@ def _replace(match: re.Match[str]) -> str:
         return self._UNICODE_ESCAPE_RE.sub(_replace, content)
 
     def _strip_sql_comments(self, content: str) -> str:
-        content = self._SQL_BLOCK_COMMENT_INNER_UPPER_RE.sub("", content)
-        content = self._SQL_BLOCK_COMMENT_INNER_LOWER_RE.sub("", content)
-        content = self._SQL_BLOCK_COMMENT_RE.sub(" ", content)
+        content = self._SQL_BLOCK_COMMENT_STRIP_RE.sub(" ", content)
         content = self._SQL_LINE_COMMENT_RE.sub(" ", content)
         return content
 

guard_core/sync/handlers/suspatterns_handler.py

@@ -26,6 +26,8 @@
 _CTX_SENSITIVE_FILE = frozenset({"url_path", "request_body", "unknown"})
 _CTX_CMS_PROBING = frozenset({"url_path", "request_body", "unknown"})
 _CTX_RECON = frozenset({"url_path", "unknown"})
+_CTX_PROTO_POLLUTION = frozenset({"query_param", "request_body", "unknown"})
+_CTX_CODE_INJECTION = frozenset({"query_param", "request_body", "unknown"})
 _CTX_ALL = frozenset({"query_param", "header", "url_path", "request_body", "unknown"})
 
 
@@ -47,6 +49,8 @@
         "sensitive_file",
         "cms_probing",
         "recon",
+        "proto_pollution",
+        "code_injection",
     }
 )
 
@@ -67,6 +71,8 @@
     "sensitive_file": _CTX_SENSITIVE_FILE,
     "cms_probing": _CTX_CMS_PROBING,
     "recon": _CTX_RECON,
+    "proto_pollution": _CTX_PROTO_POLLUTION,
+    "code_injection": _CTX_CODE_INJECTION,
 }
 
 
@@ -122,6 +128,19 @@ class SusPatternsManager:
             _CTX_SQLI,
             "sqli",
         ),
+        (r"\w/\*(?!!)[^*]*\*/\w", _CTX_SQLI, "sqli"),
+        (r"(?i)(?:OR|AND)\s+'[\w\d]*'='[\w\d]*'?", _CTX_SQLI, "sqli"),
+        (
+            r"(?i)(?:^|;)\s*(?:DROP|TRUNCATE|ALTER)\s+(?:TABLE|DATABASE|SCHEMA)\b",
+            _CTX_SQLI,
+            "sqli",
+        ),
+        (
+            r"(?i);\s*(?:INSERT\s+INTO|UPDATE\s+\w+\s+SET|DELETE\s+FROM)\b",
+            _CTX_SQLI,
+            "sqli",
+        ),
+        (r"(?i)\bORDER\s+BY\s+\d+\b", _CTX_SQLI, "sqli"),
         (r"(?:\.\.\/|\.\.\\)(?:\.\.\/|\.\.\\)+", _CTX_DIR_TRAVERSAL, "dir_traversal"),
         (
             r"(?:/etc/(?:passwd|shadow|group|hosts|motd|issue|mysql/my.cnf|ssh/"
@@ -162,6 +181,16 @@ class SusPatternsManager:
             _CTX_CMD_INJECTION,
             "cmd_injection",
         ),
+        (
+            r"[;|&]\s*(?:ls|cat|rm|id|whoami|uname|wget|curl|nc|netcat|socat|bash|sh|python|perl)\b",
+            _CTX_CMD_INJECTION,
+            "cmd_injection",
+        ),
+        (
+            r"(?i)\b(?:nc|netcat|ncat)\s+-[a-z]*e\b|/dev/tcp/\d",
+            _CTX_CMD_INJECTION,
+            "cmd_injection",
+        ),
         (
             r"(?:php|data|zip|rar|file|glob|expect|input|phpinfo|zlib|phar|ssh2|"
             r"rar|ogg|expect)://[^\s]+",
@@ -194,6 +223,11 @@ class SusPatternsManager:
             "nosql",
         ),
         (r"(?:\{\s*\$[a-zA-Z]+\s*:\s*(?:\{|\[))", _CTX_NOSQL, "nosql"),
+        (
+            r'"\$(?:where|gt|gte|lt|lte|ne|eq|regex|in|nin|all|size|exists|type|mod|options|expr|jsonSchema)"\s*:',
+            _CTX_NOSQL,
+            "nosql",
+        ),
         (
             r"(?i)filename=[\"'].*?\.(?:php\d*|phar|phtml|exe|jsp|asp|aspx|sh|"
             r"bash|rb|py|pl|cgi|com|bat|cmd|vbs|vbe|js|ws|wsf|msi|hta)[\"\']",
@@ -216,6 +250,12 @@ class SusPatternsManager:
             _CTX_TEMPLATE,
             "template",
         ),
+        (r"<%[=#]?[^%]*%>", _CTX_TEMPLATE, "template"),
+        (
+            r"\$\{[^}]*(?:@[\w.]+@|\b\w+\s*\(|\d+\s*[*/%+\-]\s*\d+)[^}]*\}",
+            _CTX_TEMPLATE,
+            "template",
+        ),
         (
             r"[\r\n]\s*(?:HTTP\/[0-9.]+|Location:|Set-Cookie:)",
             _CTX_HTTP_SPLIT,
@@ -334,6 +374,16 @@ class SusPatternsManager:
         (r"(?:^|/)autodiscover/", _CTX_RECON, "recon"),
         (r"^/dns-query(?:\?|$)", _CTX_RECON, "recon"),
         (r"(?:^|/)\.git/(?:refs|index|HEAD|objects|logs)(?:/|$)", _CTX_RECON, "recon"),
+        (
+            r"(?:__proto__|constructor)\s*(?:\[\s*[\"']prototype[\"']\s*\]|\.\s*prototype)|[\"']__proto__[\"']\s*:",
+            _CTX_PROTO_POLLUTION,
+            "proto_pollution",
+        ),
+        (
+            r"System\.Diagnostics\.Process\.Start\s*\(|System\.Reflection\.|Assembly\.Load\s*\(",
+            _CTX_CODE_INJECTION,
+            "code_injection",
+        ),
     ]
 
     patterns: list[str] = [p[0] for p in _pattern_definitions]

tests/attack_simulation/README-campaign.md

@@ -0,0 +1,36 @@
+# Attack-simulation campaign — human gate
+
+The AI-coordinated campaign (`.claude/workflows/attack-campaign.workflow.js`, untracked)
+is a discovery tool. It never edits the committed corpus. It writes proposals to
+`tests/attack_simulation/out/` (git-ignored):
+
+- `campaign-report.md` — human-readable confirmed bypasses, clustered by class+technique.
+- `campaign-proposals.json` — machine-readable, each with a `disposition` (`seed` or `transform`).
+
+The campaign's own verification is only as strong as the model it ran on. Treat every
+proposal as a candidate, not a fact. Re-verify each against the real detector before
+promoting it:
+
+```python
+import asyncio
+from tests.attack_simulation.score_payloads import score_payloads
+print(asyncio.run(score_payloads(["<candidate payload>"])))
+```
+
+A `detected: false` result confirms the detector genuinely misses it; then judge whether
+it is a genuine attack worth keeping.
+
+## Promoting a proposal (manual)
+
+1. Read `out/campaign-report.md`. Decide which confirmed bypasses are worth keeping.
+2. Re-verify each chosen payload with `score_payloads` (above) — keep only real misses.
+3. For a `seed`: add the payload to `corpus/seeds/<attack_class>.txt` under a
+   `# source: campaign-derived, verified` header line.
+4. For a `transform`: add the function to `mutations.py` and a unit test in
+   `test_mutations.py`, then register it in `TRANSFORMS`.
+5. Regenerate the baseline:
+   `uv run python -c "import asyncio,json; from pathlib import Path; from tests.attack_simulation.harness import run_benchmark; b=Path('tests/attack_simulation'); r=asyncio.run(run_benchmark(b/'corpus')); (b/'baseline.json').write_text(json.dumps({'detection_rate':r['detection_rate'],'fp_rate':r['fp_rate']},indent=2,sort_keys=True)+chr(10))"`
+6. Run `uv run pytest tests/attack_simulation/ --no-cov` and commit the corpus +
+   baseline changes. The ratchet now protects the new coverage.
+
+Nothing from the campaign is committed except what a human promotes here.

tests/attack_simulation/__init__.py

@@ -0,0 +1,23 @@
+import asyncio
+from pathlib import Path
+
+from tests.attack_simulation.harness import run_benchmark
+from tests.attack_simulation.reporter import write_reports
+
+_BASE = Path(__file__).parent
+
+
+def main() -> None:
+    report = asyncio.run(run_benchmark(_BASE / "corpus"))
+    json_path, md_path = write_reports(report, _BASE / "out")
+    print(f"wrote {json_path}")
+    print(f"wrote {md_path}")
+    print(
+        f"detection_rate={report['detection_rate']:.3f} "
+        f"fp_rate={report['fp_rate']:.3f} "
+        f"runtime_s={report['runtime_seconds']:.2f}"
+    )
+
+
+if __name__ == "__main__":
+    main()

tests/attack_simulation/__main__.py

@@ -0,0 +1,4 @@
+{
+  "detection_rate": 0.8568421052631578,
+  "fp_rate": 0.125
+}