Back to skill

Security audit

Child Poor Posture (Hunchback / Head Tilt) Real-Time Reminder | 儿童坐姿不良(驼背/歪头)实时提醒

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because scanners found privacy and credential-handling risks, plus a mismatch between its advertised live posture reminders and the submitted-video analysis and history features they observed.

Consider the verdict's Review status carefully before installing, especially if processing a child's video. The scanners report possible remote video upload, identity and token handling, and an option that may analyze video despite its environment-only name. The scan was partial because a referenced local path was missing, so some evidence is incomplete; these findings do not establish malicious intent.

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 (29)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied chunk implements user-database infrastructure, not the declared posture-monitoring behavior. Although database storage could support a larger application, this code specifically manages user identities and token fields without any visible connection to posture measurements or reminders. No camera capture, angle estimation, threshold monitoring, or audio output appears in the chunk. This is a material purpose and resource mismatch on the supplied evidence; it does not establish malicious behavior or external data transmission.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied chunk implements account provisioning, credential management, and remote API access rather than camera-based posture analysis. It contains no video capture, angle estimation, sustained-posture detection, or voice-reminder implementation. Generic formatting and polling helpers could be supporting details, but automatic remote registration, identity-file access, persistent authentication, and payment handling are material undeclared capabilities. These findings apply to the supplied chunk; they do not establish what other files implement or that these methods are invoked automatically on import.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied code does not implement the declared posture-monitoring workflow and instead exposes an unrelated, currently broken chat helper. No external agent execution actually occurs because the command and subprocess call are commented out. The imported base classes could contain additional functionality, but their behavior cannot be established from this chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied chunk implements a general-purpose API client rather than camera processing, posture-angle estimation, duration tracking, or voice reminders. Its download-link and user-directory operations have no demonstrated connection to the declared posture-monitoring purpose. Network helpers could support a larger posture application, but no such integration is shown here. This assessment is limited to the supplied chunk: it defines callable methods and does not itself demonstrate executed requests, data exfiltration, or malicious behavior.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The supplied code is primarily an API-backed video-analysis and reporting client. It processes existing files or URLs and polls for completed reports; it does not capture a camera stream or implement the declared posture-triggered feedback loop. The imported API implementation is not supplied, so posture analysis might occur remotely, but its presence and specificity cannot be verified here. Video submission and polling could support the declared service, yet the visible workflow and additional report-management capabilities materially differ from the stated real-time monitoring and voice-reminder purpose.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The visible code implements an on-demand video-analysis and history-retrieval CLI, materially different from the described real-time posture-monitoring and reminder system. Analysis is delegated to skill.get_output_analysis(), whose implementation is not supplied, so posture-specific processing inside that dependency cannot be ruled out. Nevertheless, user-linked history retrieval is an explicit additional capability, and this chunk does not demonstrate the declared live monitoring or voice-alert behavior.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
89% confidence
Finding

The wrapper's posture-related labels do not establish the advertised real-time detection and reminder behavior. The shown implementation delegates analysis and prints or saves the result, while also providing an account-associated history-listing operation. The imported skill implementation is not supplied, so it could contain some advertised analysis functionality; nevertheless, this chunk does not substantiate the central monitoring-and-reminder workflow and exposes an additional undeclared capability. Identity resolution and output-file writing alone are supporting implementation details, not independent grounds for the mismatch.

Content

No source excerpt is available for this finding.

YARA rule 'agent_skill_mcp_tool_poisoning_metadata': MCP/tool metadata poisoning indicators in tool schemas or skill manifests [agent_skills]

High
Category
YARA Match
Confidence
80% confidence
Finding

YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Content

Scanner excerpt · SKILL.md (reported line 3)May include surrounding context.

md
---
name: "smyx-child-poor-posture-detection-analysis"
description: "Using the camera built into a smart desk lamp or mounted above the desk, the system analyzes the child's sitting-posture video in real time, detecting spinal curvature angle (estimated Cobb angle) and head tilt angle. | 通过智能台灯内置摄像头或书桌上方摄像头,实时分析儿童学习时的坐姿视频,检测脊柱弯曲角度(Cobb角估算)以及头部倾斜度(侧倾角)。当驼背(Cobb角>10°)或歪头(头部侧倾角>15°)持续时间超过预设阈值(如5秒)时,触发语音提醒(如'请坐直'、'头抬正'),帮助儿童�

The HTTP helper can attach service credentials and user identity to arbitrary caller-supplied destinations.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

The helper accepts arbitrary absolute HTTP or HTTPS URLs and automatically adds access tokens, an API key, authorization data, and user identity without checking the destination origin. If an attacker can influence a URL passed to this helper, credentials and identity can be sent to their server; allowing HTTP also permits cleartext exposure.

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 supplying it still runs video analysis when an input is provided (L053–L054, L070–L082).

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The parser accepts --export-env-only and promises not to analyze, but the program never checks the flag before initializing identity and invoking analysis. A caller relying on this non-executing mode can unintentionally trigger remote processing of a child's video, making this a significant privacy and consent failure.

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 report generation rather than the advertised real-time posture monitoring and sustained-posture voice alerts.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

The manifest promises continuous camera-video analysis and voice reminders when posture thresholds remain exceeded for a configured duration. Instead, this implementation accepts a video file or URL, submits it for analysis (L110–139), and polls for a completed report at five-second intervals for up to 24 attempts; the shown workflow contains no live-camera monitoring, duration-based posture tracking, or voice-reminder dispatch.

Content

No source excerpt is available for this finding.

Local video contents are passed to the analysis API without disclosure that the file will be uploaded.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The code reads the complete local video into a multipart file payload and passes it to the analysis API without a disclosure or consent gate at this boundary. Although API usage and general guardian consent are mentioned in the documentation, it does not clearly explain that local child recordings are transferred remotely, creating a heightened risk of uninformed disclosure.

Content

No source excerpt is available for this finding.

The documented environment-only mode still executes video analysis.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The --export-env-only flag is parsed but never used to stop execution, so supplying an input still invokes video analysis despite the explicit promise not to do so. In this child-video workflow, reliance on the flag can result in unintended remote disclosure and processing of sensitive recordings.

Content

No source excerpt is available for this finding.

The service exposes username-based user lookup, which is unrelated to posture detection or voice reminders.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

At L098–L101, get_user_by_username delegates a caller-supplied username to a user-lookup utility. The manifest describes camera-based posture analysis and spoken reminders, with no account-directory or user-search functionality that would justify this capability.

Content

No source excerpt is available for this finding.

L448–L461 define persistent identity and authentication-token storage unrelated to the stated posture-monitoring purpose.

Medium
Category
Excessive Agency
Confidence
91% confidence
Finding

The model supports persisting reusable authentication tokens alongside identifying information as ordinary plaintext SQLite fields, with no explicit restrictive database permissions or secret-storage protection shown. A party able to read the database or its backups could recover credentials; linking this information to a child-monitoring service increases the privacy consequences, although identity storage itself is not inherently unrelated or malicious.

Content

No source excerpt is available for this finding.

FileUtil.open silently truncates existing files without disclosing its destructive behavior.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

The generically named open helper always uses mode 'w', which immediately truncates an existing file even before the caller writes anything. No docstring, comment, confirmation, or action log explains this behavior; the exception handler only reports failures.

Content

No source excerpt is available for this finding.

The request helper silently provisions a remote service account and persists authentication data, beyond the described posture-monitoring behavior.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

An ordinary request can trigger remote login or registration and save returned profile information and tokens locally without a separate enrollment or consent step. This creates undisclosed identity-processing and credential-retention side effects, which are especially sensitive in a child-monitoring skill; the behavior appears to be service onboarding rather than proven malicious exfiltration.

Content

No source excerpt is available for this finding.

An ordinary HTTP request can silently transmit identity information and register a remote account.

Medium
Category
Output Handling
Confidence
93% confidence
Finding

The request wrapper can submit identity information with both silent login and registration enabled as an implicit side effect. Users expecting posture analysis are not given a distinct authorization boundary for account creation, creating a privacy risk that is heightened when the service concerns children.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
89% confidence
Finding

External HTTP transmission is not inherently unsafe, but this call sends identity fields to a configured remote login endpoint during implicit registration without a visible consent check. The code does not show video transmission at this location; the supported concern is undisclosed identity disclosure, compounded by the lack of an enforced HTTPS requirement.

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 writes missing configuration files.

Low
Category
Output Handling
Confidence
94% confidence
Finding

YamlUtil.load creates parent directories and writes a YAML file when the requested path is absent, without a comment, docstring, prompt, or log disclosing this side effect. BaseEnum.init_subclass invokes this loader during class definition (L044–L059), so importing the module can modify the filesystem without an explicit save request.

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()