Back to skill

Security audit

Turtle Pneumonia Symptom (Open-Mouth Breathing) Detection | 龟类张嘴呼吸(肺炎征兆)识别

Security checks for vulnerabilities and agentic risk

Overview

This cloud video analyzer needs review because its request helper can expose service credentials and its export-only option still allows video processing.

Wait for fixes to credential destination checks and the export-only option. Selected videos or video URLs go to Life Emergence, and service use can create an account and save authentication tokens locally. Review its retention and billing terms before sharing footage; the package's retention statements do not demonstrate server-side enforcement.

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
99% confidence
Finding

The supplied chunk implements user-account database infrastructure rather than turtle respiratory-symptom detection. It contains no camera access, video processing, feeding-state classification, mouth-opening measurement, mucus detection, or pneumonia-risk alert generation. Database support alone could be an implementation detail, but this code specifically manages user identities, personal data, and tokens without an evident connection to the declared purpose. This assessment is limited to the supplied chunk; it does not establish what other skill components do 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 monitoring purpose. It provides generic scaffolding and a disabled, broken agent-chat helper. No subprocess command actually executes in this chunk. Imported base classes and API services are not supplied, so their behavior cannot be assessed; the conclusion is limited to the visible code.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The security-relevant behavior is the reported attachment of API keys and authentication tokens to caller-supplied absolute URLs: an attacker-controlled destination could receive credentials and potentially use them to access reports or consume account resources. Automatic account management and billing messages are not independently proof of malicious behavior; this conclusion relies on the static finding because the referenced implementation was not supplied for verification.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The supplied chunk contains no camera processing, mouth-opening frequency analysis, feeding-state detection, mucus or discharge detection, or pneumonia-risk reporting. Instead, it exposes general-purpose API functionality, including user lookup. HTTP helpers could support a larger monitoring application, but this chunk does not establish that connection. The assessment is limited to the supplied code; endpoint destinations, authentication behavior, and actual data transfers cannot be determined from it.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
86% confidence
Finding

Local and remote video inputs, report formatting, and saving results are reasonable supporting features of the declared analysis purpose. However, the --list command adds historical-record retrieval, and its accompanying comment specifies an undeclared automatic trigger. The snippet establishes the exposed command interface but does not verify actual symptom detection, record storage, or enforcement of the documented trigger.

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-turtle-pneumonia-symptom-detection-analysis"
description: "Through fixed enclosure cameras, the system analyzes mouth and nasal videos of turtles to detect abnormally frequent open-mouth breathing in non-feeding states (mouth opening frequency unusually elevated), as well as the presence of mucus (reflective spots or strands) or nasal discharge around the mouth and nose. | 通过龟缸固定摄像头,分析龟类的口鼻部视频,检测龟在非进食状态下(未摄食时)口部频繁开合(张嘴呼吸,频率异常增高),以及口鼻区域是否有黏液(反光点或丝状物)或鼻腔分泌物。当同时或单�

Service credentials are automatically attached to requests even when the caller supplies an unrelated absolute URL.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

The helper accepts arbitrary absolute HTTP or HTTPS URLs and attaches service tokens and the API key without checking the destination's origin. If an attacker can influence a caller's URL, credentials can be sent to an attacker-controlled server; HTTP destinations additionally expose them in plaintext. Turtle-analysis functionality does not justify sending service credentials to unrelated hosts.

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 analysis workflow uses a remote API without warning users that their video or supplied URL will be shared with that service.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

Although the document mentions cloud APIs, it directs automatic processing of local media without requiring a clear user-facing disclosure or authorization for external transfer. Enclosure recordings can reveal private surroundings, while supplied camera URLs may contain access tokens; sharing either with an external service crosses a privacy boundary, and the stated retention restrictions are not evidence of enforcement.

Content

No source excerpt is available for this finding.

The documented export-only mode still executes video analysis.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The --export-env-only flag promises not to perform analysis, but args.export_env_only is never checked: when an input or URL is supplied, execution still reaches skill.get_output_analysis(). This violates the user's explicit execution boundary and could cause unintended video processing, remote submission, or service charges, depending on the downstream implementation. Enclosure-camera footage may contain private surroundings, making unexpected processing more consequential, although transmission is not established by this file alone.

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 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 entire selected local video and passes it to the analysis API as an upload, with no disclosure or consent check visible in this file. Although remote video analysis is consistent with the skill's purpose, enclosure-camera footage may incidentally expose people, surroundings, or other private information, and users may expect local processing. The API destination, retention practices, and any disclosure elsewhere are not available here, so this supports a privacy concern rather than a conclusion of malicious exfiltration.

Content

No source excerpt is available for this finding.

The --export-env-only option promises not to perform analysis, but the code ignores it and runs analysis normally.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The --export-env-only flag explicitly promises not to execute analysis, but its parsed value is never checked, so an invocation that also supplies --input or --url proceeds to skill.get_output_analysis. This violates an explicit no-analysis boundary and could cause unintended video processing, external disclosure, or API charges depending on the downstream implementation; enclosure-camera footage may contain private surroundings, although this file alone does not establish what data is transmitted.

Content

No source excerpt is available for this finding.

The service exposes username-based user lookup, an account-directory capability unrelated to turtle symptom detection.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

At L098-L101, get_user_by_username forwards a supplied username to RequestUtil.get_user_by_username. Looking up user accounts is not a direct requirement of analyzing turtle mouth/nasal videos or producing pneumonia-risk warnings, and the manifest does not declare account-management functionality. This establishes an additional capability, not evidence that user lookups occur during analysis.

Content

No source excerpt is available for this finding.

The skill includes user-profile and credential database management unrelated to turtle symptom detection.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

L448–L465 define a system-user table containing personal details and token/open_token fields, and L470–L473 expose it through a DAO with inherited read, update, and delete capabilities. DAO initialization also creates tables and migrates sys_user (L180–L215), making this an implemented database-management capability rather than merely an unused field declaration. Managing user identities and credentials is not justified by the manifest's camera-based symptom analysis and risk-warning purpose.

Content

No source excerpt is available for this finding.

The documented agent-analysis call is disabled; ai_chat performs no analysis and returns no result.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The docstring at L26 claims that ai_chat invokes the openclaw agent through subprocess, and L33–L34 promise an analysis result or session ID. However, the invocation is commented out at L61–L66, and the replacement dictionary at L68 causes result.stderr at L70 to raise an exception that is caught without returning a result. This actively contradicts the documented analysis behavior rather than merely omitting an implementation detail.

Content

No source excerpt is available for this finding.

FileUtil.open silently truncates existing files without disclosing its destructive write mode.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

The generically named open helper always uses mode 'w', which immediately erases an existing file's contents even before anything is written. There is no docstring, explanatory comment, confirmation, or operation-specific log disclosing this behavior; exception tracing does not disclose successful truncation.

Content

No source excerpt is available for this finding.

The request helper silently registers remote accounts and persists user credentials, beyond the stated turtle-analysis purpose.

Medium
Category
Excessive Agency
Confidence
95% confidence
Finding

At L550-L561, the helper posts an identity to /sys/phoneLogin with silent=1 and register=1. When authentication tokens are missing, L580-L604 invokes this flow and saves returned user information and tokens through UserDao; automatic account provisioning and persistent identity management are not declared requirements of detecting turtle respiratory symptoms.

Content

No source excerpt is available for this finding.

HTTP requests can silently register a remote account and persist returned authentication tokens.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

When tokens are missing, the generic HTTP helper invokes a separate phoneLogin request with registration enabled, transmitting the current identifier as both openId and mobile. It can then save the returned user information and authentication tokens locally, without a confirmation, explanatory comment or docstring, or user-facing disclosure of these additional side effects.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
80% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

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 a comment, docstring, confirmation, or log disclosing this side effect. BaseEnum.init_subclass invokes this loader during class definition (L044–L059), so merely importing the module can modify the filesystem rather than only read configuration.

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 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: