veritensor 1.7.0__tar.gz → 1.8.0__tar.gz

{veritensor-1.7.0/src/veritensor.egg-info → veritensor-1.8.0}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: veritensor
-Version: 1.7.0
+Version: 1.8.0
 Summary: Antivirus for the AI Supply Chain. Scans models, datasets, notebooks, and RAG documents for threats.
 Author-email: Arsenii Brazhnyk <veritensor@gmail.com>
 License: Apache-2.0
@@ -29,6 +29,8 @@ Requires-Dist: huggingface_hub>=0.19.0
 Requires-Dist: cyclonedx-python-lib>=6.0.0
 Requires-Dist: jinja2>=3.1.0
 Requires-Dist: pillow>=12.2.0
+Requires-Dist: lxml>=6.1.0
+Requires-Dist: gitpython>=3.1.47
 Provides-Extra: test
 Requires-Dist: pytest>=7.0.0; extra == "test"
 Requires-Dist: pytest-mock>=3.10.0; extra == "test"
@@ -77,7 +79,8 @@ Unlike standard SAST tools (which focus on code), Veritensor understands the bin
 2.  **Data & RAG:** Streaming scan of **Parquet, CSV, Excel, PDF** to detect Data Poisoning, Prompt Injections, and PII.
 3.  **Notebooks:** Hardening of **Jupyter (.ipynb)** files by detecting leaked secrets (using Entropy analysis), malicious magics, and XSS.
 4.  **Supply Chain:** Audits **dependencies** (`requirements.txt`, `poetry.lock`) for Typosquatting and known CVEs (via OSV.dev).
-5.  **Governance:** Generates cryptographic **Data Manifests** (Provenance) and signs containers via **Sigstore**.
+5.  **Agentic AI & MCP Servers:** Pure AST analysis of Python files to detect Agent Hijacking risks in `@mcp.tool()` functions. Scans `claude_desktop_config.json` and `mcp.json` for over-privileged permissions.
+6.  **Governance:** Generates cryptographic **Data Manifests** (Provenance) and signs containers via **Sigstore**.
 
 ---
 
@@ -183,6 +186,24 @@ Create a standalone, interactive HTML dashboard of your scan results:
 veritensor scan ./project --html
 ```
 
+### 11. Scan MCP Servers for Agent Hijacking
+Detect dangerous tool logic and over-privileged permissions in your AI agent infrastructure:
+```bash
+# Scan MCP server Python files (AST analysis — no code execution)
+veritensor scan ./mcp_servers/
+
+# Also audit MCP configuration files
+veritensor scan ./claude_desktop_config.json
+```
+
+**Example output:**
+```
+CRITICAL: MCP Agent Hijacking Risk [OS_COMMAND_EXECUTION] in tool 'run_script'
+  (line 14): os.system() inside agent tool — no human-in-the-loop confirmation
+HIGH: MCP Config [LETHAL_TRIFECTA] server 'everything':
+  filesystem + network + private data — prompt injection can silently exfiltrate all data
+```
+
 **Example Output:**
 ```Text
 ╭────────────────────────────────╮
@@ -325,7 +346,11 @@ For custom parsers and SOAR automation:
 ```bash
 veritensor scan ./models --json
 ```
-
+### 5. Excel report
+Create an Excel report with data about the problems:
+```bash
+veritensor scan ./models --excel
+```
 ---
 
 ## 🔐 Supply Chain Security (Container Signing)
@@ -403,6 +428,8 @@ veritensor_audit:
   script:
     - veritensor scan . --jobs 4
   allow_failure: false
+```
+
 ---
 
 ## 📂 Supported Formats

{veritensor-1.7.0 → veritensor-1.8.0}/README.md

@@ -17,7 +17,8 @@ Unlike standard SAST tools (which focus on code), Veritensor understands the bin
 2.  **Data & RAG:** Streaming scan of **Parquet, CSV, Excel, PDF** to detect Data Poisoning, Prompt Injections, and PII.
 3.  **Notebooks:** Hardening of **Jupyter (.ipynb)** files by detecting leaked secrets (using Entropy analysis), malicious magics, and XSS.
 4.  **Supply Chain:** Audits **dependencies** (`requirements.txt`, `poetry.lock`) for Typosquatting and known CVEs (via OSV.dev).
-5.  **Governance:** Generates cryptographic **Data Manifests** (Provenance) and signs containers via **Sigstore**.
+5.  **Agentic AI & MCP Servers:** Pure AST analysis of Python files to detect Agent Hijacking risks in `@mcp.tool()` functions. Scans `claude_desktop_config.json` and `mcp.json` for over-privileged permissions.
+6.  **Governance:** Generates cryptographic **Data Manifests** (Provenance) and signs containers via **Sigstore**.
 
 ---
 
@@ -123,6 +124,24 @@ Create a standalone, interactive HTML dashboard of your scan results:
 veritensor scan ./project --html
 ```
 
+### 11. Scan MCP Servers for Agent Hijacking
+Detect dangerous tool logic and over-privileged permissions in your AI agent infrastructure:
+```bash
+# Scan MCP server Python files (AST analysis — no code execution)
+veritensor scan ./mcp_servers/
+
+# Also audit MCP configuration files
+veritensor scan ./claude_desktop_config.json
+```
+
+**Example output:**
+```
+CRITICAL: MCP Agent Hijacking Risk [OS_COMMAND_EXECUTION] in tool 'run_script'
+  (line 14): os.system() inside agent tool — no human-in-the-loop confirmation
+HIGH: MCP Config [LETHAL_TRIFECTA] server 'everything':
+  filesystem + network + private data — prompt injection can silently exfiltrate all data
+```
+
 **Example Output:**
 ```Text
 ╭────────────────────────────────╮
@@ -265,7 +284,11 @@ For custom parsers and SOAR automation:
 ```bash
 veritensor scan ./models --json
 ```
-
+### 5. Excel report
+Create an Excel report with data about the problems:
+```bash
+veritensor scan ./models --excel
+```
 ---
 
 ## 🔐 Supply Chain Security (Container Signing)
@@ -343,6 +366,8 @@ veritensor_audit:
   script:
     - veritensor scan . --jobs 4
   allow_failure: false
+```
+
 ---
 
 ## 📂 Supported Formats

{veritensor-1.7.0 → veritensor-1.8.0}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
 
 [project]
 name = "veritensor"
-version = "1.7.0"
+version = "1.8.0"
 description = "Antivirus for the AI Supply Chain. Scans models, datasets, notebooks, and RAG documents for threats."
 readme = "README.md"
 requires-python = ">=3.10"
@@ -34,7 +34,9 @@ dependencies = [
     "cyclonedx-python-lib>=6.0.0",
     "jinja2>=3.1.0",
     "pillow>=12.2.0",
+    "lxml>=6.1.0",
     # Fickling removed from here to avoid installing PyTorch (800MB+) for regular users
+    "gitpython>=3.1.47",
 ]
 
 

veritensor-1.8.0/src/veritensor/__init__.py

@@ -0,0 +1,2 @@
+__version__ = "1.8.0"
+

{veritensor-1.7.0 → veritensor-1.8.0}/src/veritensor/cli/main.py

@@ -45,7 +45,7 @@ from veritensor.reporting.sarif import generate_sarif_report
 from veritensor.reporting.sbom import generate_sbom
 from veritensor.reporting.manifest import generate_manifest
 from veritensor.reporting.html_report import generate_html_report
-
+from veritensor.reporting.excel_report import generate_excel_report
 from veritensor.integrations.enterprise_scanner import EnterpriseScanner
 from veritensor.engines.static.mcp_scanner import scan_mcp_server
 
@@ -324,7 +324,12 @@ def scan_worker(args: Tuple[str, VeritensorConfig, Optional[str], bool, bool, bo
                 if mcp_result.mcp_tools_found:
                     for t in mcp_result.to_threat_strings():
                         scan_res.add_threat(t)
-
+        elif ext == ".json" and file_path:
+            if is_mcp_config_file(file_path):
+                from veritensor.engines.static.mcp_permission_auditor import audit_mcp_config
+                perm_result = audit_mcp_config(file_path)
+                for t in perm_result.to_threat_strings():
+                    scan_res.add_threat(t)
         else:
             # If the format is unknown to any engine at all.
             # We add INFO, but DO NOT call add_threat(),
@@ -509,6 +514,7 @@ def scan(
     api_key: Optional[str] = typer.Option(None, envvar="VERITENSOR_API_KEY", help="API Key for reporting"),
     verbose: bool = typer.Option(False, "--verbose", "-v", help="Show detailed logs"),
     html_output: bool = typer.Option(False, "--html", help="Generate a beautiful HTML report"),
+    excel_output: bool = typer.Option(False, "--excel", help="Generate an Excel report for auditors (.xlsx)"),
     output_file: Optional[str] = typer.Option(None, "--output-file", "-o", help="Save machine-readable output (JSON/SARIF/SBOM) to a file instead of stdout"),
 ):
     """
@@ -593,6 +599,21 @@ def scan(
         if not is_machine_output:
             console.print(f"\n[bold green]✅ HTML Report saved to: {report_path}[/bold green]")        
 
+    if excel_output:
+        try:
+            excel_path = generate_excel_report(filtered_results)
+            if not is_machine_output:
+                console.print(f"\n[bold green]✅ Excel Report saved to: {excel_path}[/bold green]")
+        except ImportError as e:
+            console.print(f"[bold yellow]⚠️  Excel export requires openpyxl: pip install veritensor[excel][/bold yellow]")
+    # Usage:
+    #   veritensor scan ./models/ --excel
+    #   veritensor scan ./models/ --excel --output-file report.xlsx
+    #
+    # Note: --output-file with --excel saves to the specified path.
+    # To support custom path, replace the generate_excel_report call with:
+    #   excel_path = generate_excel_report(filtered_results, output_path=output_file or "veritensor-report.xlsx")
+
     machine_text = None        
     if sarif_output:
         machine_text = generate_sarif_report(results)

{veritensor-1.7.0 → veritensor-1.8.0}/src/veritensor/core/streaming.py

@@ -39,13 +39,21 @@ class RemoteStream(io.IOBase):
             parsed = urlparse(url)
             if parsed.scheme not in ("http", "https"):
                 raise ValueError(f"Invalid scheme: {parsed.scheme}")
-            
+
             domain = parsed.netloc.lower()
             is_allowed = (domain in ALLOWED_DOMAINS) or (domain.endswith(".huggingface.co"))
-            
+
+            # warning-only allowed any URL through,
+            # completely bypassing the domain allowlist.
             if not is_allowed:
-                logger.warning(f"Security Warning: Accessing external domain: {domain}")
-                
+                raise ValueError(
+                    f"SSRF Protection: Domain '{domain}' is not in the allowed list "
+                    f"({', '.join(ALLOWED_DOMAINS)}). "
+                    f"Only HuggingFace domains are permitted for remote streaming."
+                )
+
+        except ValueError:
+            raise
         except Exception as e:
             raise ValueError(f"Invalid URL format: {e}")
 

{veritensor-1.7.0 → veritensor-1.8.0}/src/veritensor/engines/static/dependency_engine.py

@@ -149,8 +149,19 @@ def _check_osv_batch(packages: Dict[str, str]) -> List[str]:
     """Queries OSV.dev API in a single batch request."""
     threats = []
     payload = {"queries": []}
-    pkg_list = [] 
-    
+    pkg_list = []
+
+    # OSV.dev does not publish a hard batch limit, but large payloads risk
+    # 413 / rate-limiting. Cap at 500 packages per call to stay well within limits.
+    # For most real-world lock files (poetry.lock, Pipfile.lock) this is sufficient.
+    OSV_BATCH_LIMIT = 500
+    if len(packages) > OSV_BATCH_LIMIT:
+        logger.warning(
+            f"Dependency list has {len(packages)} packages — "
+            f"truncating OSV check to first {OSV_BATCH_LIMIT} entries."
+        )
+        packages = dict(list(packages.items())[:OSV_BATCH_LIMIT])
+
     for name, version in packages.items():
         payload["queries"].append({
             "package": {"name": name, "ecosystem": "PyPI"},
@@ -292,3 +303,5 @@ def _safe_read_text(path: Path) -> str:
         except UnicodeDecodeError:
             continue
     return raw_bytes.decode('utf-8', errors='ignore')
+
+