Back to skill

Security audit

Driver Head-Pose Abnormality (Head-Down / Side-View) | 驾驶员头部姿态异常(低头/侧视)检测

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because scanners report a mismatch between its advertised live driver-monitoring purpose and its video submission workflow, along with unexpected identity handling and other side effects.

Consider the verdict suspicious and review the reported data handling before installing. In particular, consider whether you are comfortable submitting driver videos to a remote service and allowing identity or account information to be sent and stored. Do not rely on this skill for live driving alerts based on the scan evidence. The AIG scan reported no findings, but the Skillspector analysis was partial because a referenced local path was missing.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
Findings (26)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied chunk is an agent/API utility layer rather than a driver-monitoring implementation. Generic helpers alone would not establish a mismatch, but remote account registration, identity-file access, persistent user/token management, and account-balance handling are material capabilities not disclosed by the camera-based safety description. No visible code connects these capabilities to driver monitoring. This assessment is limited to the supplied chunk and does not establish malicious intent or rule out a separate monitoring implementation elsewhere.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied chunk implements a general-purpose user database layer, not driver head-pose monitoring. It contains no camera acquisition, pitch/yaw computation, distraction threshold or duration evaluation, or warning/voice-reminder logic. Although database utilities can support a larger application, the user-profile and token-management functionality shown has no demonstrated connection to the declared DMS purpose. This assessment is limited to the supplied code; it does not establish what other files implement or imply malicious intent.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The visible code does not implement the declared camera input, pitch/yaw calculations, angle thresholds, two-second persistence checks, or voice-linked alerts. Its agent-chat helper is nonfunctional as written and does not actually execute an external command. Imported base classes and API services are not supplied, so their behavior cannot be assessed; this finding is limited to the provided chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The visible code provides general backend API functionality rather than the declared driver-monitoring behavior. Networking could support such a system, but this chunk establishes no connection between its storage, user-directory, or generic API operations and head-pose analysis or distraction alerts. This assessment is limited to the supplied chunk; it does not establish that other parts of the skill lack the declared functionality or that these methods are executed.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The real-time monitoring and warning claims exceed the demonstrated submitted-video and report-polling implementation, creating a safety-relevant reliance risk. History listing and export links are explicitly documented, however, and are not themselves undeclared behavior.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The documented live DMS and voice-warning behavior is not supported by the visible file/URL submission workflow. That mismatch can mislead safety-sensitive deployments, although identity-associated report retrieval is disclosed in the workflow and is not independently suspicious.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The skill advertises continuous real-time monitoring and voice warnings, while the supplied workflow submits individual videos for remote analysis. In a driving-safety context, relying on that claim could leave distraction events without timely warnings; backend analysis does not establish a live alerting capability.

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, file_read, file_write, network, shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

The --export-env-only option promises not to execute analysis, but the program ignores it and proceeds with video analysis.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The export-only flag is parsed but never checked, so combining it with a video input still initiates analysis and potentially uploads driver footage. Identity initialization also occurs despite the promised non-analysis mode, violating the user's expectation of a side-effect-free configuration operation.

Content

No source excerpt is available for this finding.

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · skills/smyx_analysis/scripts/skill.py (reported line 28)May include surrounding context.

python
result_json = JsonUtil.parse(result_json_pure_text, result_json_pure_text)

        result_json_common_ai_response = result_json.get("commonAiResponse") if isinstance(result_json,
                                                                                           dict) else result_json
        if result_json_common_ai_response:
            result_json = result_json_common_ai_response

Whitespace Padding

Medium
Category
Prompt Injection
Confidence
70% confidence
Finding

Large whitespace padding was detected (a block of blank lines or a long run of spaces). This can push injected instructions below or to the right of the visible area so a human reviewer never sees them while the agent still reads them. Manual review of the hidden content is recommended.

Content

Scanner excerpt · skills/smyx_analysis/scripts/skill.py (reported line 33)May include surrounding context.

python
result_json = result_json_common_ai_response

        result_json_health_ai_response = result_json.get("healthAiResponse") if isinstance(result_json,
                                                                                           dict) else result_json
        if result_json_health_ai_response:
            result_json = result_json_health_ai_response

The skill implements submitted-video analysis and report polling rather than the advertised real-time DMS monitoring workflow.

Medium
Category
Excessive Agency
Confidence
95% confidence
Finding

This implementation submits a file or URL and polls for a report rather than continuously monitoring a camera and delivering timely warnings. Even if the remote service estimates head pose correctly, the visible workflow does not support the advertised real-time safety behavior.

Content

No source excerpt is available for this finding.

Local video contents are submitted to the analysis API without disclosure that the file leaves the local system.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The code reads the entire local video and passes it to the remote analysis API without an upload-specific consent check. Although cloud/API use and driver consent are mentioned elsewhere, local-file analysis is not clearly identified as transferring sensitive facial footage to an external processor.

Content

No source excerpt is available for this finding.

The documented export-only mode does not prevent video analysis.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The export-only option is never used to control execution, so a caller supplying it alongside a video still triggers remote analysis. This can transmit sensitive driver footage despite an explicit request for configuration output only, and identity initialization is also unconditional.

Content

No source excerpt is available for this finding.

The skill includes persistent user-profile and authentication-token management unrelated to head-pose monitoring.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The DAO supports storing authentication tokens alongside identifiable profile data in an ordinary workspace SQLite database, without visible encryption or explicit restrictive file permissions. Unrelated functionality alone is not a vulnerability, but persisting reusable credentials this way increases exposure to backups, other workspace readers, and local compromise without an evident requirement from head-pose monitoring.

Content

No source excerpt is available for this finding.

ai_chat is documented to invoke an analysis agent but never executes it and silently returns None.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The docstring at L26 and L33–L34 promises subprocess-based invocation of the openclaw agent and an analysis result or session ID. However, subprocess execution is commented out at L61–L66, and the empty dictionary at L68 causes result.stderr at L70 to raise AttributeError, which is caught at L81–L82 without returning a result. This contradicts the method's documented analysis behavior.

Content

No source excerpt is available for this finding.

FileUtil.open silently truncates existing files without disclosing its write-only behavior.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

The generically named open helper always opens the supplied path in 'w' mode, immediately truncating an existing file. No docstring, explanatory comment, confirmation, or user-facing message discloses this destructive behavior; exception tracing only runs after a failure.

Content

No source excerpt is available for this finding.

HTTP requests can silently register a remote account and persist its authentication credentials, beyond the stated driver-monitoring purpose.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

An ordinary HTTP request can automatically initiate remote registration, obtain authentication credentials, and persist them through UserDao without an explicit enrollment step. This creates undisclosed identity and credential side effects beyond the stated monitoring function, including local exposure of persisted tokens.

Content

No source excerpt is available for this finding.

HTTP requests can silently trigger remote account registration and transmit the current user identity.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

The helper explicitly submits silent registration flags and sends the current identity as both openId and mobile when credentials are missing. Because this happens inside a general request method rather than a disclosed enrollment action, using the monitoring skill can cause unexpected identity transmission and account creation.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
96% confidence
Finding

This POST transmits an identity value to the configured registration service, and its caller prioritizes ApiEnum.API_SECRET_KEY as that identity, allowing the secret to be copied into both openId and mobile. The actual destination is not shown, so malicious ownership cannot be inferred, but sending a secret as ordinary account metadata risks disclosure through service storage and logging independently of any legitimate authentication use.

Content

Scanner excerpt · skills/smyx_common/scripts/util.py (reported line 561)May include surrounding context.

python
"source": ConstantEnum.DEFAULT__SKILL_HUB_NAME
            }
            try:
                _response = requests.post(_url, json=_data)
                if _response.status_code == 200:
                    _response_json = _response.json()
                    if _response_json and _response_json.get("success"):

Configuration loading silently creates directories and files, including during module import.

Low
Category
Output Handling
Confidence
90% confidence
Finding

YamlUtil.load creates parent directories and writes a YAML file when the requested configuration is missing, without any explanatory comment, docstring, or user-facing notice. BaseEnum.init_subclass invokes this loader during class definition (L044–L059), so importing the module can modify the filesystem unexpectedly.

Content

No source excerpt is available for this finding.

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · skills/smyx_common/scripts/dao.py (reported line 243)May include surrounding context.

python
username = getattr(model, "username", None)
            if username:
                column_names = self.__model__.__table__.columns.keys()
                update_data = {key: getattr(model, key) for key in column_names if key != "username"}
                return self.update_by_username(username, **update_data)

            return None

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · skills/smyx_common/scripts/dao.py (reported line 353)May include surrounding context.

python
column_names = self.__model__.__table__.columns.keys()

            for key in column_names:
                value = getattr(model, key)
                setattr(instance, key, value)

            session.commit()

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · skills/smyx_common/scripts/dao.py (reported line 326)May include surrounding context.

python
if filters:
                for key, value in filters.items():
                    query = query.filter(getattr(self.__model__, key) == value)

            if offset:
                query = query.offset(offset)

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · skills/smyx_common/scripts/dao.py (reported line 441)May include surrounding context.

python
if filters:
                for key, value in filters.items():
                    query = query.filter(getattr(self.__model__, key) == value)

            return query.scalar()
        finally: