Back to skill

Security audit

Pet Body Condition Score 3D Analysis | 宠物体态3D评分(BCS)

Security checks for vulnerabilities and agentic risk

Overview

Scanners classify this skill as suspicious because they report undeclared account and report-history features, credential and identity handling, and an option that can still run analysis when it promises not to.

Consider the reported account creation, identity and token storage, report-history access, and export-only behavior before installing. Skillspector completed most analysis but marked coverage partial because two local references could not be resolved, so the evidence is incomplete; the scanners do not establish malicious intent.

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 (1)

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/smyx_pet_body_condition_score_3d_analysis.py
Finding

Default initialization discards the upstream sender identity and shares report access

Content
View full analysis

Impact

A sender can receive reports created by another sender through this fallback account. New analyses are also attributed to the shared account instead of the supplied sender identity.

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 (27)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The video-analysis commands and output options align with the declared purpose. However, the --list command and explicitly documented history-related triggers introduce a separate report-history capability not disclosed in the description. The supplied chunk does not show the module implementation, so its server-side analysis, data storage, and report-access mechanisms cannot be verified.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The supplied chunk is a general user-account persistence component, not a pet-video analysis implementation. Database persistence can be a supporting implementation detail, but this code specifically manages personal identity and credential records rather than videos, body reconstructions, BCS scores, or observation results. The declared purpose does not describe this account-management capability. No video processing, analysis API calls, or external transmission are visible in this chunk; the finding is limited to its implemented database behavior.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
90% confidence
Finding

The primary video-analysis workflow matches the description at the wrapper level, but history retrieval is an additional user-facing capability that can run without any video input. It is distinct from analyzing a supplied video. The delegated implementation is not included, so this chunk alone cannot verify the claimed 3D reconstruction, BCS scoring, or server-side processing. Local result saving is a supporting implementation detail, not a mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

The main video-analysis workflow matches the declared purpose. However, --list invokes get_output_analysis_list(open_id=...) and exits without requiring video input, exposing a distinct historical-record retrieval capability not described in the declaration. Identity resolution and output-file writing are supporting implementation details, not independent mismatches. The imported analysis implementation is not supplied, so actual 3D reconstruction, BCS scoring, and remote API behavior cannot be verified from this chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

Ordinary HTTP transport, polling, and authentication can support the declared server-side analysis and are not inherently mismatches. However, this chunk goes beyond basic transport: it silently provisions remote accounts, can transmit the configured API secret as account-identifying fields, and adds authentication headers even when the caller supplies an arbitrary absolute URL. These identity-management and credential-sharing behaviors are not represented by the pet-analysis description. The payment-installation response is an additional undeclared workflow; it returns instructions rather than actually installing anything. No pet-video analysis is implemented in this utility chunk, but its absence alone is not the basis for the finding.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
80% confidence
Finding

The HTTP wrappers, pagination, and download-URL helper are plausible supporting infrastructure and are not independently mismatches. However, the code also exposes a dedicated username-based user lookup, an account-related capability outside the declared purpose, with no visible connection to video analysis or authentication. The missing pet-analysis logic in this shared helper alone does not establish a mismatch. The lookup implementation is not supplied, so its destination and returned data cannot be determined.

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-pet-body-condition-score-3d-analysis"
description: "Triggers when a user provides multi-angle pet videos (standing/side/top-down view) URL or files for analysis; supports local video uploads or network URLs to call server-side APIs for 3D body shape reconstruction and Body Condition Score (BCS, 1-9) evaluation, classifying body type as underweight, ideal, or overweight/obese, outputting standardized observation data (without diagnosing diseases or prescribing treatment). Application scenarios: smart feeders, pet cameras, pet health management platforms. Development reason: early warning for obesity-related diseases, scientific wei

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 run analysis, but the code still invokes it.

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. When an input is supplied, execution still initializes identity and invokes analysis, potentially uploading private household video and consuming service credits 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
88% confidence
Finding

The code packages the entire local file and passes it to the analysis API, but provides no confirmation, log, or comment explaining that the file is transmitted to a service. A request to analyze a local video does not necessarily imply consent to remote processing, particularly if the video contains private information.

Content

No source excerpt is available for this finding.

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

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The environment-export-only argument is parsed but never used to control execution. Supplying it alongside a video still invokes identity resolution and analysis, potentially transmitting private footage and incurring charges contrary to the option's explicit no-analysis promise.

Content

No source excerpt is available for this finding.

The service exposes username-based user lookup unrelated to pet body-condition analysis.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

At L098-L101, get_user_by_username forwards a supplied username to a user-lookup utility. Looking up user accounts is a separate capability from uploading pet videos, reconstructing body shape, and retrieving BCS observations, and the manifest does not declare an account-lookup purpose.

Content

No source excerpt is available for this finding.

The save() method is documented as creating a new record but can overwrite an existing user after an insertion failure (L221-L244).

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

Any insertion exception can trigger an update by ID and then by username or realname, copying fields including identifiers and authentication tokens into an existing user. If an untrusted record reaches this method, a collision can become an unintended credential or identity overwrite; the source does not establish an externally reachable account-takeover path.

Content

No source excerpt is available for this finding.

ai_chat documents agent execution but never invokes an agent or produces the documented analysis result.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The docstring at L25–L34 promises to invoke openclaw agent through subprocess and return analysis output or a session ID. However, the command and subprocess invocation are commented out; the replacement empty dictionary at L68 causes the attribute access at L70 to fail, and the exception handler suppresses the error, leaving the method to return None. This contradicts the documented analysis delegation rather than merely changing its implementation.

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
97% confidence
Finding

The generic open helper always uses mode 'w', which immediately erases an existing file's contents. No docstring, comment, confirmation, or user-facing log explains that this helper opens files for replacement rather than reading.

Content

No source excerpt is available for this finding.

Analysis requests can silently register a remote user account and persist its credentials, beyond the declared video-analysis workflow.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

An ordinary analysis request can initiate silent registration using the current identity and persist returned authentication tokens in the local database. Server-side analysis reasonably requires authentication, but undisclosed account creation and credential retention exceed merely submitting a video and create privacy and credential-exposure risks.

Content

No source excerpt is available for this finding.

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

Medium
Category
Output Handling
Confidence
91% confidence
Finding

The HTTP helper sends the current identity as both openId and mobile while requesting silent registration, without an approval step in the shown flow. This is the same undisclosed provisioning issue identified by SDI-1, not an independent additional exploit.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
84% confidence
Finding

The transmitted payload contains the current identity and requests registration at the configured health-service endpoint. External API traffic is expected for this skill, but this particular transmission creates an undisclosed identity-provisioning side effect; there is no evidence that the endpoint 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"):

Generic history-report phrases can automatically trigger the pet BCS report lookup.

Low
Category
Trigger Abuse
Confidence
90% confidence
Finding

The examples include generic history phrases as automatic triggers, while the instructions require an immediate cloud lookup. In an unrelated reporting conversation, this can cause an unintended external request and disclosure of pet-analysis history into the conversation without establishing that the user meant this service.

Content

No source excerpt is available for this finding.

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

Low
Category
Output Handling
Confidence
92% confidence
Finding

YamlUtil.load creates parent directories and writes a YAML file when the requested configuration is missing, without a log, prompt, docstring, or explanatory comment disclosing this side effect. BaseEnum.init_subclass invokes this loader during class definition (L044–L059), so importing the module can modify the filesystem even though no save operation 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)