Back to skill

Security audit

Employee Emotion Fluctuation HR Report | 员工入职/离职情绪波动预警(HR方向)

Security checks for vulnerabilities and agentic risk

Overview

This skill needs review because its anonymity claims conflict with identifiable employee reports, and broken controls can send footage against the user's selected settings.

Avoid using real employee footage until the destination and no-analysis controls are fixed, cloud processing is explicitly authorized, and consent and opt-outs are enforced. Workstation-linked reports should be treated as identifiable employee data. Require transparent outreach and protected credential storage, and independently verify the service's retention and HR access controls; the bundled code does not establish those protections.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/smyx_employee_emotion_fluctuation_hr_analysis.py
Finding

Ignored API endpoint sends sensitive video to the default service

Content
View full analysis

Impact

Raw employee footage, or the supplied video URL including any embedded access parameters, is transmitted to the default service instead of the selected service. This is unintended disclosure, not evidence that the default service itself is malicious.

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/smyx_employee_emotion_fluctuation_hr_analysis.py
Finding

No-analysis flag does not prevent cloud processing

Content
View full analysis

Impact

Employee footage or its source URL can be submitted for cloud analysis despite the user's explicit instruction not to analyze it. Identity initialization can also perform local state changes before any such restriction is checked.

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
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Findings (28)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The visible implementation materially differs from the declared purpose: it contains no camera processing or employee-monitoring workflow, and its chat wrapper is nonfunctional because subprocess execution is disabled and dictionary attributes are accessed incorrectly. Imported base and API-service implementations are not supplied, so their capabilities cannot be assessed. No actual subprocess execution or camera access is demonstrated by 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 account and API infrastructure rather than employee-camera monitoring. It contains no camera acquisition, expression or posture analysis, consent or anonymization enforcement, historical-baseline computation, or HR alert generation. Generic utilities alone would not establish a mismatch, but the concrete remote account-provisioning and credential-bearing request capabilities are not represented in the declaration. This assessment is limited to the supplied chunk; other files could implement the declared monitoring functions.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied chunk’s concrete purpose is general user-account database management, materially different from the declared employee-monitoring workflow. Database CRUD alone could be supporting infrastructure, but this implementation specifically manages identity and credential fields without any visible connection to the claimed monitoring or anonymization capabilities. This assessment is limited to the supplied chunk; other components may implement the declared workflow.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The privacy-relevant mismatch is that the documented workflow submits office footage or URLs for analysis without an explicit pre-submission consent and anonymization gate, despite presenting those safeguards as foundational. Missing local baseline calculations alone would not establish a vulnerability because analysis could occur remotely, but unverified safeguards can expose identifiable employee footage.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

The actionable issue is the gap between promised consent/anonymity and a workflow that proceeds directly from supplied footage to API analysis. On-demand rather than continuous processing is not intrinsically unsafe, and absent client-side analytics do not prove the remote service lacks them.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
82% confidence
Finding

The generic AI-analysis endpoints could support the declared system and are not inherently mismatched. However, the explicit health-order report export endpoint introduces an unrelated resource and capability absent from the description. This finding is limited to endpoint configuration: the chunk does not demonstrate actual report retrieval or export. Missing monitoring functionality alone is not evidence of mismatch because this is only a configuration file.

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-employee-emotion-fluctuation-hr-analysis"
description: "Using fixed cameras in enterprise office areas (with employee consent and anonymization), the system performs long-term monitoring of employees' facial expressions and posture features, building per-person historical baselines (smile frequency, sigh count, frown level, etc.). | 通过企业办公区固定摄像头(需征得员工同意并匿名化处理),长期监测员工的面部表情和姿态特征,建立个人历史基线(如笑容频率、叹气次数、皱眉程度等)。当检测到某员工近期的笑容频率显著下降(例如比基线降低40%)、叹�

The analysis workflow lacks a clear disclosure about sending sensitive office footage to the analysis API.

High
Category
Output Handling
Confidence
95% confidence
Finding

Although API use is mentioned, the workflow does not clearly disclose before processing that identifiable office footage or its URL will be transferred to an external analysis service. General consent to workplace monitoring does not establish consent to that transfer, particularly when the skill repeatedly promises anonymization.

Content

No source excerpt is available for this finding.

Anonymity and voluntary-care framing culminates in identifying monitored employees and concealing the basis for targeted outreach.

High
Category
Prompt Injection
Confidence
99% confidence
Finding

The document calls outputs anonymous while explicitly directing managers to identify employees from workstation coordinates. Requiring outreach to conceal its monitoring basis further undermines meaningful transparency, and the resulting identifiable emotion profiles could enable discrimination or unauthorized employment use despite written prohibitions.

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 default trigger selects employee-emotion monitoring for any office-camera video analysis request.

Medium
Category
Trigger Abuse
Confidence
98% confidence
Finding

The default trigger treats any office-camera analysis request as authorization for employee emotion monitoring. A user requesting an unrelated task could therefore initiate sensitive employee profiling and remote processing without expressly requesting it.

Content

No source excerpt is available for this finding.

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

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The parser accepts --export-env-only and promises no analysis, but the program never checks the flag. When supplied with an input or URL, it continues into video analysis, potentially transferring sensitive employee footage despite the caller explicitly selecting a non-analysis mode.

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 an analysis API without disclosure of the upload.

Medium
Category
Output Handling
Confidence
98% confidence
Finding

The code reads the complete local video into a multipart file payload and passes it to the analysis API without a visible disclosure, authorization check, or de-identification step. In this employee-monitoring context, that can transfer recognizable faces and workplace activity beyond the privacy expectations established by the skill's anonymity claims.

Content

No source excerpt is available for this finding.

The --export-env-only option promises not to perform analysis, but does not prevent analysis from running.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The --export-env-only flag is parsed but never used to stop execution, so supplying a video still invokes the analysis workflow. This violates an explicit no-analysis expectation and can cause unintended remote processing of sensitive workplace footage.

Content

No source excerpt is available for this finding.

save() is documented as creating a new record but silently falls back to overwriting existing records.

Medium
Category
Excessive Agency
Confidence
97% confidence
Finding

Any insertion exception causes save() to attempt updates, including a fallback that matches either username or realname and copies all remaining columns. A failed create can therefore overwrite another matching record's identity or authentication fields instead of reporting failure, compromising account-data integrity.

Content

No source excerpt is available for this finding.

FileUtil.open silently truncates existing files.

Medium
Category
Output Handling
Confidence
97% confidence
Finding

The generically named open method always uses write mode, which immediately erases an existing file's contents. No docstring, explanatory comment, confirmation, or user-facing log discloses this destructive behavior; exception tracing does not warn about successful truncation.

Content

No source excerpt is available for this finding.

The request helper silently provisions remote user accounts, an undeclared capability beyond employee emotion analysis.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

An ordinary request can automatically call a remote login endpoint with registration enabled and persist the returned authentication tokens locally. This undeclared account-provisioning side effect exports identity information without a separate authorization step; consent to employee monitoring does not establish consent to remote account creation.

Content

No source excerpt is available for this finding.

HTTP requests can silently trigger remote account registration.

Medium
Category
Output Handling
Confidence
95% confidence
Finding

The request helper sets both 'silent' and 'register' before posting to the remote phoneLogin endpoint, so a request can create persistent remote account state without explicit caller approval. This is particularly concerning for a workplace-monitoring skill because identifiers can become associated with an external account beyond the stated analysis workflow.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
94% confidence
Finding

The registration payload transmits the supplied username as both openId and mobile to a configured remote service, without a separate disclosure or consent gate. When API_SECRET_KEY supplies current__user_name, the code can also place that credential into account-identity fields, exposing it to profile storage and identity-processing logs rather than keeping it confined to authentication.

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"):

Loading configuration silently creates directories and files, including during module import.

Low
Category
Output Handling
Confidence
88% confidence
Finding

When a configuration file is absent, YamlUtil.load creates its parent directory and writes a YAML file without a warning, log, comment, or docstring explaining this side effect. BaseEnum.init_subclass invokes this loader at L053 and L058, so defining the enum classes during import can modify the filesystem even though no explicit save was requested.

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)