Back to skill

Security audit

Fish Gasping & Ammonia Poisoning Visual Warning | 水族箱内氨氮中毒视觉预兆(鱼浮头)

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because its aquarium-monitoring purpose is accompanied by broad account, data, and remote-service behavior, including a reported risk of sending a configured secret during registration.

Consider the verdict and the incomplete scan evidence before installing. In particular, review whether remote video processing, account creation, identity or credential handling, and the export-only behavior are acceptable for your use; the supplied results do not establish the behavior of a missing referenced path or the host runtime's permission 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
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
Findings (30)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied chunk is a user-oriented database utility, not an implementation of aquarium-camera analysis. It contains no camera ingestion, fish detection, mouth or gill movement analysis, multi-fish duration tracking, or risk-warning logic. Database persistence can be a legitimate supporting detail, but this chunk specifically manages user identities, personal attributes, and token fields rather than fish observations or alerts. That materially different functionality supports a mismatch finding for the supplied code; it does not establish malicious behavior or rule out fish-analysis functionality elsewhere.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Generic formatting, polling, and filesystem helpers could support the declared skill and are not themselves a mismatch. However, the account provisioning, credential handling, identity-bearing network requests, and billing workflow are substantive capabilities not represented in the aquarium-monitoring description. This chunk contains no camera processing, fish-behavior detection, multi-fish duration tracking, or water-quality warning logic. The finding is limited to the supplied chunk; it does not establish whether other files implement the declared monitoring functionality or whether these callable utilities are actually invoked.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The visible implementation does not substantiate the declared primary purpose. It provides generic class scaffolding and a broken chat helper, not fish-behavior monitoring. The commented-out agent command does not execute and should not be counted as an active capability. Imported base classes and API-service implementations were not supplied, so their behavior cannot be established from this chunk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The supplied code implements an API-backed video-analysis/report client rather than the described aquarium-monitoring logic. It contains no visible camera integration, fish-behavior detection, 60-second symptom threshold, or specified warning generation. Those features could exist in the imported API service or remote backend, but the supplied chunk does not establish that connection. Report-history retrieval is an additional user-facing capability absent from the description. File validation, API submission, and polling are supporting implementation details and are not independently treated as mismatches.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

The supplied chunk implements a reusable API client rather than camera processing, fish-behavior detection, duration tracking, or risk alerts. HTTP helpers could support the declared system, but this chunk also exposes user lookup and general resource-management capabilities with no demonstrated connection to that purpose. The mismatch assessment is limited to the supplied code; it does not establish malicious behavior or show that these methods are invoked.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
87% confidence
Finding

The history-listing operation is an additional user-facing capability, not merely an implementation detail of analyzing fish behavior. The supplied chunk otherwise delegates analysis to skill.get_output_analysis; the imported implementation is unavailable, so the absence of visible fish detection or 60-second threshold logic does not establish that those advertised features are missing. The mismatch finding is based on the explicit, undeclared history-retrieval capability.

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-gasping-ammonia-warning-analysis"
description: "Through fixed aquarium cameras, the system analyzes fish behavior near the water surface, detecting repeated mouth-out-of-water (gasping), rapid mouth opening/closing (fast respiration) and exaggerated operculum (gill cover) movement — classic symptoms of hypoxia or poisoning. | 通过鱼缸固定摄像头,分析鱼类在水面附近的行为,检测鱼嘴反复探出水面(浮头)、张口快速开合(类似喘气)、鳃盖运动加剧等缺氧或中毒典型症状。当多条鱼同时出现上述行为且持续时间超过设定阈值(默认 60 秒)时,输出

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 lacks an explicit privacy warning that submitted aquarium media or URLs are processed by a remote API.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The workflow invokes remote analysis without requiring a clear user-facing disclosure that local footage or supplied URLs leave the environment. Although API use is mentioned elsewhere, aquarium cameras can capture private interiors, people, or access-bearing camera URLs, making undisclosed transfer a meaningful privacy risk.

Content

No source excerpt is available for this finding.

Emergency alerts notify all household/management contacts without requiring a user-facing privacy warning.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The documented emergency policy broadcasts alerts to all household or management contacts without specifying recipient-level opt-in or a privacy notice. General deployment authorization does not necessarily authorize broad disclosure of tank incidents, and false detections could cause unwanted repeated notifications.

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 advertises --export-env-only as not performing analysis, but the parsed flag is never checked. With an input supplied, execution still initializes identity and invokes analysis, potentially uploading private footage and consuming service credits despite the user's explicit no-analysis selection.

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
94% confidence
Finding

The implementation reads the complete local media file and passes it to the analysis API without a disclosure or authorization gate in this path. Footage from aquarium cameras can include private surroundings or people, so treating local-file analysis as implicit permission for remote transfer risks unintended disclosure.

Content

No source excerpt is available for this finding.

The --export-env-only option promises not to analyze video, but does not prevent analysis.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The --export-env-only flag is accepted but never used to stop execution. When media input is supplied, the program still performs identity initialization and remote analysis, violating the promised no-analysis mode and potentially causing unwanted data transfer or charges.

Content

No source excerpt is available for this finding.

The service exposes generic remote mutation capabilities unrelated to fish-behavior analysis or risk warnings.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

The add, edit, delete, http_put, and http_delete methods forward caller-supplied URLs and arguments to HTTP request utilities (L050–L095). General-purpose remote record modification and deletion are not justified by the manifest's camera-analysis and warning scope; this finding concerns those mutation capabilities, not network access needed for camera integration or alerts.

Content

No source excerpt is available for this finding.

Username-based user lookup is outside the declared aquarium-monitoring purpose.

Medium
Category
Excessive Agency
Confidence
86% confidence
Finding

The get_user_by_username method delegates an arbitrary supplied username to a user-lookup utility (L098–L101). The manifest describes analyzing fish behavior and issuing water-quality warnings, but does not establish a need for general username-based user lookup.

Content

No source excerpt is available for this finding.

The DAO implements personal-profile and credential management unrelated to aquarium behavior analysis.

Medium
Category
Excessive Agency
Confidence
93% confidence
Finding

The User model stores real names, email addresses, birthdays, demographics, and token/open_token credentials (L448–L465), and UserDao exposes CRUD operations for these records (L470–L473). DAO initialization also creates and migrates the user table (L184–L215), so this is implemented database functionality rather than unused field documentation. Managing these personal and credential records is not justified by the manifest's camera-based fish-distress detection and warning purpose.

Content

No source excerpt is available for this finding.

ai_chat claims to invoke an analysis agent, but its implementation performs no agent invocation.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The agent invocation is commented out, and accessing stderr on an empty dictionary raises an exception that is swallowed, causing the method to return None. If the monitoring workflow relies on this method, analysis and time-sensitive fish-distress warnings can silently disappear; the supplied source does not establish that the active workflow calls it.

Content

No source excerpt is available for this finding.

FileUtil.open silently truncates existing files without disclosing its write-only behavior.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

Opening a path with mode 'w' immediately erases existing contents, even before the caller writes anything. This generic open helper has no docstring, explanatory comment, confirmation, or action-specific log disclosing that destructive behavior; exception tracing only runs after failure.

Content

No source excerpt is available for this finding.

The HTTP helper silently provisions remote user accounts and persists authentication credentials, beyond the stated fish-monitoring purpose (L550–L604).

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

An ordinary HTTP request can implicitly initiate remote login or registration, transmit an identity, and cache returned tokens in a local database without an explicit enrollment step. Authentication may be legitimate for hosted analysis, but hiding these identity and credential side effects inside a general request helper exceeds the disclosed fish-monitoring behavior.

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
95% confidence
Finding

The helper sets register=1 and silent=1 and submits the current identity automatically when authentication tokens are missing. Users requesting aquarium analysis are not given a separate opportunity to authorize account creation or the associated identity disclosure.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
94% confidence
Finding

The POST transmits the selected identity to the configured health-service endpoint without destination or HTTPS validation. More seriously, the surrounding code prioritizes API_SECRET_KEY as that identity, so a configured secret can be copied into both openId and mobile and exposed to registration processing, logs, or an insecure endpoint.

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"):

The user-facing analysis heading is fixed to Chinese without a language choice.

Low
Category
Output Handling
Confidence
92% confidence
Finding

The output-heading method unconditionally returns Chinese text, imposing that language on this part of the response. The file provides no user opt-in, locale selection, or documented justification for the restriction.

Content

No source excerpt is available for this finding.

Configuration loading silently creates directories and writes missing configuration files.

Low
Category
Output Handling
Confidence
93% confidence
Finding

YamlUtil.load creates parent directories and writes a YAML file when the requested path is absent, without a prompt, log, docstring, or explanatory comment. Because BaseEnum.init_subclass calls this loader during class definition (L044–L059), importing the module can modify the filesystem even though configuration loading would ordinarily be expected to read existing data.

Content

No source excerpt is available for this finding.