Back to skill

Security audit

Fish Feeding Behavior Activity Analysis | 鱼类摄食行为活跃度分析

Security checks for vulnerabilities and agentic risk

Overview

The skill is marked suspicious because scanners report undeclared account and credential handling, risky data transfers, and a mismatch between the fish-analysis description and parts of the supplied implementation.

Treat this as a review warning and consider avoiding installation until the reported credential handling, account registration, video transfer disclosures, export-only behavior, and cross-user history exposure are clarified or addressed. Skillspector completed scans across the listed components but marked its analysis partial because a referenced path was missing; imported components and behavior beyond the supplied material therefore remain unassessed. The suspicious verdict indicates concerning behavior or weaknesses, not proof that the author intended harm.

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_fish_feeding_activity_analysis.py
Finding

Default CLI identity resolution discards the upstream sender identity

Content
View full analysis

Impact

After one sender submits an analysis under the fallback account, another sender sharing the workspace can retrieve that account's history, including analysis content and generated report links. This is an ordinary authorization defect, not evidence of a backdoor.

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 (28)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The visible code does not implement the declared primary purpose. Its only substantive helper is a nonfunctional generic agent-chat wrapper, not a feeding-analysis pipeline. The commented subprocess code does not execute and is not evidence of active command execution. Imported base classes and services are not supplied, so their capabilities cannot be assessed; this conclusion is limited to the provided chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Database persistence can be a supporting implementation detail, but the concrete functionality here centers on user identities and tokens rather than feeding observations or analysis results. The supplied chunk contains no camera capture, object detection, motion analysis, feed estimation, scoring, or appetite-warning implementation. The mismatch is the unrelated account-management capability, not merely the use of SQLite. No network transmission or token exfiltration is shown, and this assessment is limited to the supplied chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

No camera capture, video analysis, fish detection, feeding-score calculation or appetite-warning logic appears in this chunk. Generic utilities and HTTP transport could support the declared feature and would not alone establish a mismatch. However, the module additionally implements an undeclared identity and account lifecycle: it reads data/smyx-api-key.txt, creates a persistent default user when needed, calls /sys/phoneLogin with silent registration enabled, and saves returned tokens and user information. These behaviors access resources and perform account operations not represented in the fish-monitoring description. The assessment is limited to this chunk; it does not establish that fish-analysis functionality is absent elsewhere or that the observed network activity is malicious.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
91% confidence
Finding

Networking and download helpers could support the declared functionality, so their presence alone would not establish a mismatch. However, this chunk exposes a broader generic API-management interface, including user lookup, with no demonstrated connection to the stated purpose. The assessment is limited to the supplied code: it defines these capabilities but does not show them being invoked, and it does not establish malicious activity.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
84% confidence
Finding

The analysis entry point is consistent with the declared purpose, although the supplied chunk delegates processing and does not establish whether the promised fish metrics, scoring, or warnings are implemented. The separate history-listing operation accesses user-associated records and is not disclosed in the description. Local/URL input handling and optional result-file output are supporting implementation details, not mismatches. No specific upload behavior or remote destination can be established from this chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
80% confidence
Finding

The generic AI-analysis routes could support the declared purpose and are not inherently mismatched. However, the explicit health-order report-export resource introduces an unrelated integration not explained by the description. This finding is limited to configuration: the chunk does not itself issue requests or export reports. The absence of fish-analysis logic alone would not establish a mismatch in a configuration-only 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-fish-feeding-activity-analysis"
description: "Through built-in cameras of smart feeders or fixed cameras on aquariums, the system captures fish feeding videos after feeding. Using AI object detection and motion analysis, it identifies the number of fish gathering for food, feeding intensity (fish swimming speed, feeding action frequency), and remaining feed amount, and computes a comprehensive feeding activity score (0-100). | 通过智能喂食器内置摄像头或鱼缸固定摄像头,在投喂后拍摄鱼群摄食视频,利用 AI 目标检测和运动分析技术,识别鱼群聚集抢食的数量、摄食强度(�

The request helper forwards application credentials to caller-supplied absolute URLs without restricting the destination.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

The helper accepts arbitrary absolute HTTP or HTTPS URLs and automatically attaches application tokens, the API key, and user-identifying metadata. A caller able to influence the destination can send these secrets to an attacker-controlled server; allowing HTTP also permits cleartext credential 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 local-video analysis workflow lacks a clear warning that camera footage is sent to a remote API.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The instructions direct analysis of local camera files without clearly telling users that their footage will be uploaded to an external processor. General references to APIs and cloud history do not adequately disclose this transfer, and aquarium footage may also reveal people, home interiors, or commercially sensitive facilities.

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 parser accepts --export-env-only and explicitly promises not to analyze, but execution never checks the flag. With an input supplied, it still initializes identity and invokes the analysis workflow, potentially transferring footage or consuming service resources contrary to the user's selected 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
96% confidence
Finding

The local-file branch reads the complete video into a multipart file payload and passes it to the analysis service without a visible upload notice or authorization check. Camera footage can contain private surroundings or facility information even when its intended subject is fish.

Content

No source excerpt is available for this finding.

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

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The --export-env-only flag is parsed but never used to stop execution. A caller supplying footage under the explicit promise that analysis will not run can instead trigger identity initialization and the analysis service, creating an unintended data-transfer or resource-consumption risk.

Content

No source excerpt is available for this finding.

L098-L101 expose user-account lookup by username, a capability unrelated to fish-feeding activity analysis.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

The manifest describes analyzing feeding videos, computing activity scores, and reporting reduced appetite. The get_user_by_username method instead delegates to a user-lookup API, exposing an account-information capability with no stated role in that workflow.

Content

No source excerpt is available for this finding.

The DAO implements user-profile and credential management unrelated to fish-feeding analysis.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

The User model stores personal information and authentication tokens, and UserDao exposes the inherited read, update, and delete operations for those records. DAO initialization also creates registered tables and migrates sys_user (L184-L215), so this is an implemented account-management capability, not merely an unused field declaration. The manifest describes video-based feeding analysis and appetite alerts, with no stated purpose requiring management of user identities or credentials.

Content

No source excerpt is available for this finding.

ai_chat claims to invoke an analysis agent, but execution is disabled and the method cannot return an analysis result.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The docstring at L26 states that this method invokes openclaw agent through subprocess and returns an analysis result or session ID (L33–L34). However, the subprocess call is commented out, and the replacement empty dictionary at L68 causes an AttributeError at L70; the exception is caught and the method implicitly returns None without performing any analysis.

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 opens its target in 'w' mode, which immediately erases existing contents. There is no docstring, explanatory comment, confirmation, or operation log disclosing that behavior; the exception handler only handles failures.

Content

No source excerpt is available for this finding.

An analysis request can silently register a remote user and persist account credentials, exceeding the manifest's fish-feeding analysis scope.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

An ordinary request can automatically invoke a remote login endpoint with registration enabled, then persist returned account tokens in SQLite without an explicit enrollment step. This introduces undisclosed identity transmission and durable account state beyond the described fish-feeding analysis, with no consent gate visible in this path.

Content

No source excerpt is available for this finding.

HTTP authentication silently transmits the current identity to a remote login endpoint with registration enabled.

Medium
Category
Output Handling
Confidence
95% confidence
Finding

The authentication helper sends the selected identity as both openId and mobile while requesting silent registration. That identity can originate from an API secret or local identity configuration, so routine analysis requests can disclose sensitive identifiers without a visible user approval step.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
92% confidence
Finding

External requests are not inherently vulnerabilities, but this specific POST transmits the current identity and requests remote registration as an implicit side effect of another request. In the fish-analysis context, the supplied code provides no explicit consent boundary for that disclosure; this is the same underlying enrollment issue identified by the other registration findings.

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

At L017–L020, YamlUtil.load creates parent directories and writes a YAML file when the requested path does not exist, without a comment, docstring, log, or confirmation disclosing this side effect. The subclass initialization hook calls this loader at L053 and L058, 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()