fix: P2+ security bugs — zip-slip, unbounded request body, query length limit

query_intelligence/contracts.py

@@ -7,6 +7,7 @@
 
 MIN_RETRIEVAL_TOP_K = 1
 MAX_RETRIEVAL_TOP_K = 100
+MAX_QUERY_LENGTH = 2000
 
 QuestionStyle = Literal["fact", "why", "compare", "advice", "forecast"]
 TimeScope = Literal[
@@ -134,7 +135,7 @@ class RetrievalResult(BaseModel):
 
 
 class AnalyzeRequest(BaseModel):
-    query: str
+    query: str = Field(min_length=1, max_length=MAX_QUERY_LENGTH)
     user_profile: dict[str, Any] = Field(default_factory=dict)
     dialog_context: list[dict[str, Any]] = Field(default_factory=list)
     debug: bool = False
@@ -147,7 +148,7 @@ class RetrievalRequest(BaseModel):
 
 
 class PipelineRequest(BaseModel):
-    query: str
+    query: str = Field(min_length=1, max_length=MAX_QUERY_LENGTH)
     user_profile: dict[str, Any] = Field(default_factory=dict)
     dialog_context: list[dict[str, Any]] = Field(default_factory=list)
     top_k: int = Field(default=20, ge=MIN_RETRIEVAL_TOP_K, le=MAX_RETRIEVAL_TOP_K)

query_intelligence/external_data/sync.py

@@ -168,13 +168,31 @@ def _allow_patterns_for_source(source_id: str) -> list[str] | None:
     return None
 
 
+def _safe_zip_extractall(archive: zipfile.ZipFile, extract_dir: Path) -> None:
+    """Extract a zip archive while blocking zip-slip path traversal attacks.
+
+    zipfile.extractall() does not validate member paths; a malicious archive
+    could contain entries like ``../../etc/cron.d/evil`` and write files
+    outside *extract_dir*.  This helper resolves every member path and raises
+    ``ValueError`` if the resolved path escapes the target directory.
+    """
+    resolved_root = extract_dir.resolve()
+    for member in archive.infolist():
+        target = (extract_dir / member.filename).resolve()
+        if not target.is_relative_to(resolved_root):
+            raise ValueError(
+                f"Blocked unsafe zip member path (zip-slip): {member.filename!r}"
+            )
+        archive.extract(member, extract_dir)
+
+
 def _extract_zip_files(target_dir: Path) -> None:
     for path in target_dir.rglob("*.zip"):
         extract_dir = path.with_suffix("")
         if extract_dir.exists():
             continue
         with zipfile.ZipFile(path) as archive:
-            archive.extractall(extract_dir)
+            _safe_zip_extractall(archive, extract_dir)
 
 
 def _extract_7z_files(target_dir: Path) -> None:

scripts/llm_response.py

@@ -1246,6 +1246,9 @@ def coerce_query(value: Any) -> str:
     return query
 
 
+_MAX_REQUEST_BODY_BYTES = 10 * 1024 * 1024  # 10 MB – prevent OOM via huge Content-Length
+
+
 def run_service(args: argparse.Namespace) -> None:
     from http import HTTPStatus
     from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
@@ -1272,6 +1275,11 @@ def _read_json_body(self) -> dict[str, Any]:
             content_length = int(self.headers.get("Content-Length") or "0")
             if content_length <= 0:
                 return {}
+            if content_length > _MAX_REQUEST_BODY_BYTES:
+                raise ValueError(
+                    f"Request body too large: {content_length} bytes "
+                    f"(max {_MAX_REQUEST_BODY_BYTES})"
+                )
             raw = self.rfile.read(content_length)
             parsed = json.loads(raw.decode("utf-8"))
             if not isinstance(parsed, dict):