Back to skill

Security audit

Infant Stool Color Abnormality (Clay-Pale / Bloody) | 婴儿大便颜色识别(陶土色/血便)

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because scanners report behavior beyond its stated infant stool-image analysis purpose, along with privacy, consent, and unexpected-processing concerns.

Treat the suspicious verdict as a reason to avoid installing unless you can verify the remote service, what data is sent, and how account creation and token storage work. Also confirm that image processing requires your consent and that export-only mode does not process inputs; the scan was partial, so it cannot resolve behavior involving the missing reference.

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 (27)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied chunk is centered on user identity and account storage, not the declared stool-image assessment workflow. Database persistence can be a supporting implementation detail, but this code specifically manages personal profiles and authentication tokens without any visible connection to images, stool observations, classification results, or medical alerts. No camera access, visual inference, color classification, or warning generation appears in the chunk. This assessment is limited to the supplied code; it does not establish what other skill files implement or imply malicious behavior.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Generic file, JSON, date, polling, and HTTP helpers could support the declared feature and are not themselves evidence of mismatch. However, the chunk also implements an undeclared account-provisioning and billing workflow: /sys/phoneLogin is called with silent=1 and register=1, user credentials are persisted through UserDao, and HTTP 402 responses produce payment-skill installation and recharge instructions. These account and payment capabilities are materially outside the stated image-analysis and caregiver-warning purpose. This assessment is limited to the supplied chunk; it does not establish whether stool-analysis functionality exists elsewhere or imply malicious intent.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied chunk does not implement the declared medical-image analysis purpose. Its only substantive method is an inactive agent-chat wrapper, not an image-processing workflow. No subprocess actually runs because execution is commented out. Imported base classes and services could contain additional functionality, but their behavior is not supplied and cannot establish that the declared purpose is implemented.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

The supplied code explicitly presents itself as a video-analysis tool and includes an identity-linked history-listing operation. Neither is described in the declared stool-image classification workflow. No stool-color classification or medical-risk alert logic is visible in this chunk. Because analysis is delegated to an unavailable implementation, its underlying subject matter cannot be determined; the mismatch finding rests on the visible video-oriented interface and undeclared history retrieval, not on assumptions about that implementation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
86% confidence
Finding

The image inputs, report formatting, and output-file options are consistent supporting features of the declared analysis purpose. However, retrieving historical reports is an additional user-facing capability beyond analyzing an image and issuing risk reminders. The supplied content also advertises history-related automatic triggers despite an empty declared trigger list. These triggers are related to the subject matter, not evidence of an unrelated purpose. No implementation is provided to verify the actual classifier, storage mechanism, or medical-alert behavior.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
85% confidence
Finding

The supplied code implements a general media-analysis and report-management interface rather than an explicitly stool-color-specific workflow. It treats network inputs as video URLs and exposes a separate report-history listing operation. No visible logic selects stool-color analysis, classifies the stated colors, or generates the promised abnormal-color medical alerts. Those medical functions could exist in the unavailable analysis service, so their absence here alone is not conclusive; however, the visible video-input and stored-report retrieval capabilities extend beyond the declared description.

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-infant-stool-color-abnormality-analysis"
description: "Using a fixed camera above the baby-changing table or a smartphone, the system captures high-resolution images of the diaper area (or the stool itself), and uses AI visual analysis to identify stool color: normal yellow / yellow-green, abnormal clay-pale (white/clay-like, suggesting biliary obstruction), bright red (lower-GI bleeding), dark red / tarry black (upper-GI bleeding), etc. | 通过婴儿护理台上方固定摄像头或手机拍摄尿不湿区域(或直接拍摄排泄物)的高清图像,利用AI视觉分析技术识别大便颜色,包括正常黄色/黄绿色�

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 documented export-only mode still performs image analysis.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The parser advertises --export-env-only as performing no analysis, but the parsed flag is never checked before identity initialization or analysis dispatch. When combined with --input or --url, this can process sensitive infant imagery despite the caller explicitly selecting a non-analysis mode, violating the expected consent boundary.

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

Local video contents are submitted to the analysis API without disclosure of the upload.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The local-input branch reads the entire file and passes its bytes to the analysis API without a disclosure or consent check at this boundary. Although the broader skill mentions cloud services and guardian consent, it does not clearly establish informed approval for transferring the actual infant imagery to a remote processor; these images can contain particularly sensitive minor-related information.

Content

No source excerpt is available for this finding.

The skill exposes account-linked analysis history retrieval beyond the stated stool-color assessment purpose.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

show_analyze_list retrieves previous analyses using an open_id, and the --list branch at L077-L081 exposes this operation without requiring a new image or video. Retrieving account-linked historical records is a separate data-access capability not described in, or necessary for, assessing the supplied stool image and reporting abnormal colors.

Content

No source excerpt is available for this finding.

The documented export-only option does not prevent analysis.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The --export-env-only option promises not to execute analysis, but its value is never consulted after parsing. Supplying it alongside an input still initializes identity and calls the analysis workflow, potentially transferring sensitive video or imagery contrary to the caller's explicit selection.

Content

No source excerpt is available for this finding.

The service exposes username-based user lookup, a capability unrelated to stool-image analysis or risk alerts.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

At L098–L101, get_user_by_username delegates to a user-lookup utility. Looking up users by username is not a direct requirement of the manifest's image-based stool-color classification and medical-risk notification purpose, and no account-management capability is declared.

Content

No source excerpt is available for this finding.

The skill includes user-account and credential storage capabilities unrelated to stool-color analysis.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

The user model stores token and open_token as ordinary strings in a workspace SQLite database, without application-level credential protection shown. Account support is not inherently malicious or necessarily unrelated, but readable database copies or backups could expose reusable credentials; the infant-health context increases the sensitivity of associated access.

Content

No source excerpt is available for this finding.

The ai_chat docstring promises an agent invocation and analysis result, but the implementation never invokes the agent.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The docstring at L26 explicitly says the method calls the openclaw agent through subprocess, and L33-L34 promise an analysis result or session ID. However, the command and subprocess invocation are commented out; L68 instead assigns an empty dictionary, whose stderr access raises an exception that L81-L82 suppress, causing the method to return None without performing analysis. This contradicts the documented analysis behavior and could leave callers without the expected assessment.

Content

No source excerpt is available for this finding.

FileUtil.open silently truncates existing files.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The generic open helper always uses mode 'w', which immediately erases an existing file's contents even before the caller writes anything. No docstring, explanatory comment, confirmation, or user-facing log discloses this destructive behavior; the exception handler only runs if opening fails.

Content

No source excerpt is available for this finding.

The request helper silently provisions a remote user account and persists account credentials, beyond the advertised stool-color analysis.

Medium
Category
Excessive Agency
Confidence
95% confidence
Finding

An ordinary request can automatically invoke remote login or registration and save returned credentials without an explicit enrollment step in this flow. This creates undisclosed identity transmission and credential-retention risks, particularly for a service handling infant-health information; it does not establish deliberate exfiltration.

Content

No source excerpt is available for this finding.

HTTP requests can silently trigger remote account registration.

Medium
Category
Output Handling
Confidence
94% confidence
Finding

The helper sends register=1 and silent=1 to the remote login endpoint as an implicit side effect of making a request. This can create an account and disclose an identifier without a distinct user-approved enrollment action, which is especially problematic in a health-oriented skill.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
87% confidence
Finding

The POST automatically transmits the selected identifier as both openId and mobile to the configured health service; the selection can even originate from API_SECRET_KEY. Network access itself is legitimate, but this undisclosed identity transmission and possible reuse of a secret as account metadata create privacy and credential-exposure risks, without evidence that the destination is attacker-controlled.

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 configuration files, including during module import.

Low
Category
Output Handling
Confidence
94% confidence
Finding

YamlUtil.load creates parent directories and writes a YAML file when the requested configuration is absent, without a prompt, log, docstring, or explanatory comment disclosing this behavior. BaseEnum.init_subclass invokes this loader during class definition (L053 and L058), so merely 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)