Back to skill

Security audit

Fruit Ripeness Grading | 番茄/草莓果实成熟度分级

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with high confidence because scanners report substantial behavior beyond fruit-ripeness grading, including account and credential handling, broad network actions, and analysis that may run despite an export-only option.

Consider avoiding installation until the purpose and account behavior are clarified, credential handling is restricted, and media uploads and export-only behavior are made clear. The scan was marked partial because a referenced local path was missing; some imported or downstream behavior also could not be established, so the evidence is incomplete.

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_fruit_ripeness_grading_analysis.py:60
Finding

Default CLI workflow discards upstream sender identity and reuses a shared account

Content
View full analysis

Impact

A later sender can receive historical reports created by another sender under the shared fallback account. New analyses are also attributed to that account. This is a client-side identity-selection defect; it does not require a server-side authorization bypass.

Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • 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
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (32)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied code implements user-account persistence rather than fruit-image analysis or ripeness grading. Database infrastructure can be a supporting implementation detail, but this chunk specifically manages personal user data and token fields, which are not described or connected to the declared fruit-grading workflow. No image processing, visual measurement, ripeness standards, or grading logic appears here. This assessment is limited to the supplied chunk; it does not establish that these methods are invoked elsewhere or that any data is transmitted externally.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

Generic formatting, filesystem helpers, and HTTP transport could support the declared vision task and are not independently evidence of a mismatch. However, this chunk also implements an undeclared account, identity, authentication, and billing workflow, including reading a local API-key-named file and transmitting the resolved identity to a remote login/registration service. Its request wrapper also attaches credentials without restricting absolute URLs to the configured service. These capabilities materially exceed the stated fruit-ripeness functionality. They execute when the corresponding helpers are called, not necessarily on import. The supplied chunk contains no vision implementation, but that absence alone would not establish a mismatch because this is a shared utility module.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The CLI identifies itself as a video-analysis tool, documents MP4 inputs, and exposes a history-list operation using an open ID. These observable capabilities differ materially from the declared fruit-image grading workflow. The underlying get_output_analysis implementation is not supplied, so its exact analysis cannot be established and fruit grading cannot be ruled out inside that dependency. Nevertheless, this chunk explicitly exposes video analysis and history retrieval without showing a connection to the declared ripeness standards. Identity resolution and file output alone are supporting details, not independent reasons for the mismatch.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied code does not substantiate the declared primary functionality. The apparent external agent invocation is disabled, so it should not be treated as an executed command or an active external capability. Imported base classes and API services are not supplied, so their behavior cannot be assessed; this finding is limited to the visible chunk. No malicious behavior is evident.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
87% confidence
Finding

The visible functionality is a generic backend API client rather than a fruit-ripeness analyzer. HTTP helpers could legitimately support the declared skill, and their presence alone would not establish a mismatch. However, user lookup and general resource-management capabilities extend beyond the declared purpose without a visible task-related connection. This assessment is limited to the supplied chunk: the methods only act when invoked, and no malicious activity or data exfiltration is established.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
87% confidence
Finding

The supplied code implements a generic analysis-service client: remote inputs become a videoUrl parameter, results are extracted from commonAiResponse or healthAiResponse, and additional methods list previous reports and generate export links. No visible logic or API parameters identify tomatoes, strawberries, the declared visual measurements, or ripeness grades. File validation, API submission, and polling could support the declared purpose, but the exposed workflow is materially broader and differently oriented. Imported API implementations are not supplied, so fruit-specific processing in the backend cannot be ruled out.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
83% confidence
Finding

Generic AI-analysis endpoint configuration could support the declared purpose, and the absence of image-processing logic in a configuration-only chunk is not itself a mismatch. However, DETAIL_EXPORT_URL explicitly targets a health/order report-export API, which is inconsistent with the declared agricultural purpose. This chunk only configures that endpoint; it does not demonstrate an actual request, data export, or exfiltration.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
82% confidence
Finding

The pet-specific configuration and pet-health listing interface are outside the declared fruit-ripeness purpose. These are executable options, not merely unused imports. However, the underlying analysis implementation is not supplied, so this chunk does not establish whether the main analysis actually grades fruit or analyzes pets. Delegating vision processing and writing results to a file are normal supporting implementation details and are not themselves mismatches.

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-fruit-ripeness-grading-analysis"
description: "AI-powered fruit ripeness grading for tomatoes / strawberries. From smart grow-boxes or mobile phone images, uses AI vision to detect fruit color (green / light green / orange / red / dark red), colored-area ratio, gloss, and relative fruit size (against a reference object), and outputs a ripeness grade (Mature-Green / Turning / Ripe / Over-Ripe) based on preset standards. Helps growers identify the optimal harvest window and ensures flavor and shelf quality. Scenarios: smart grow-boxes, greenhouses, home vegetable gardens, fruit & vegetable cooperatives. | 通过智能种植箱�

The HTTP helper can send shared platform credentials to arbitrary absolute URLs.

High
Category
Excessive Agency
Confidence
99% confidence
Finding

The helper accepts arbitrary absolute HTTP or HTTPS URLs and automatically attaches shared access tokens, an API secret, and an authorization token without checking the destination. If an untrusted URL reaches this helper, an attacker-controlled server can receive reusable platform credentials; HTTP destinations also expose them to network interception.

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-media analysis workflow lacks a privacy warning about external API processing.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The local-media workflow does not clearly tell users that their files will be uploaded for external processing, even though API use is mentioned elsewhere. Grow-box and phone imagery can contain people, location clues, or proprietary growing conditions, so users may authorize analysis without understanding the disclosure.

Content

No source excerpt is available for this finding.

The documented export-only mode still performs analysis.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The parser advertises --export-env-only as not performing analysis, but the program never checks this flag. With an input supplied, execution still initializes identity and invokes analysis, potentially uploading private media or incurring charges 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 passed to the analysis API without disclosure of the upload.

Medium
Category
Output Handling
Confidence
91% confidence
Finding

The implementation reads the entire selected local media file and passes its contents to the analysis API without a visible disclosure or authorization check in this path. Although uploading fruit imagery serves the intended function, incidental personal or commercially sensitive content may also leave the user's environment.

Content

No source excerpt is available for this finding.

The documented export-only option does not prevent analysis.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The export-only flag is parsed but never checked, so supplying it together with an input still invokes analysis. This violates an explicit no-analysis boundary and can cause unintended processing, remote submission, or charges, depending on the downstream implementation.

Content

No source excerpt is available for this finding.

The skill exposes account-associated video-analysis history retrieval beyond the stated fruit-grading purpose.

Medium
Category
Excessive Agency
Confidence
90% confidence
Finding

The --list branch retrieves and prints analysis history associated with the current Open ID through show_analyze_list (L044-L047). Accessing historical account records is a separate capability from grading supplied fruit images, and the manifest does not describe history retrieval or account-record access.

Content

No source excerpt is available for this finding.

The service exposes general-purpose remote mutation capabilities beyond fruit-ripeness analysis.

Medium
Category
Excessive Agency
Confidence
86% confidence
Finding

The http_put and http_delete methods forward caller-supplied URLs and arguments to the request utility, providing general-purpose remote modification and deletion capabilities. Network access could reasonably support hosted image analysis, but unrestricted mutation methods are not justified by the stated purpose of grading fruit images and returning ripeness results. This file does not demonstrate that these methods are actually invoked.

Content

No source excerpt is available for this finding.

The documented create operation can silently overwrite an existing record.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

Although the separate create method is insert-only, save catches every insertion exception and then attempts to overwrite a record by ID or username/realname. An insertion failure can therefore become an unintended update of account fields, including credentials, without an explicit update request or ownership check in this layer.

Content

No source excerpt is available for this finding.

Fruit-grading support code implements unrelated user-account and credential persistence.

Medium
Category
Excessive Agency
Confidence
96% confidence
Finding

The User model stores personal information and authentication tokens, and UserDao exposes inherited account CRUD operations. Dao initialization also creates tables and migrates sys_user (L180-L215), making this operational account-management functionality rather than just an unused data definition. Managing user identities and credentials is not justified by the manifest's image-based fruit ripeness grading purpose.

Content

No source excerpt is available for this finding.

ai_chat claims to invoke an analysis agent but never executes it or returns analysis results.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The docstring at L26 states that this method invokes the openclaw agent through subprocess, and L33-L34 promise an analysis result or session ID. However, the subprocess call is commented out, and the empty dictionary assigned at L68 causes result.stderr to raise an exception that is caught and suppressed, leaving the method to return None without performing analysis.

Content

No source excerpt is available for this finding.

FileUtil.open silently truncates existing files.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

The generically named open method opens the supplied path in write mode, immediately erasing any existing contents. There is no confirmation, explanatory comment or docstring, or user-facing notice disclosing this destructive behavior; exception reporting does not disclose successful truncation.

Content

No source excerpt is available for this finding.

The request helper silently registers platform accounts and persists user credentials, beyond the declared fruit-grading workflow.

Medium
Category
Excessive Agency
Confidence
94% confidence
Finding

An ordinary HTTP operation can implicitly request account registration and persist returned user information and tokens. The fruit-grading interface does not make these account-creation and credential-retention side effects explicit, so users can incur persistent identity changes without knowingly requesting them.

Content

No source excerpt is available for this finding.

HTTP requests can silently trigger remote user registration.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

The authentication fallback sends silent=1 and register=1 to a login endpoint, allowing a routine request to create a remote account. This is a persistent external side effect that is not conveyed by the generic HTTP-helper interface or the declared grading task.

Content

No source excerpt is available for this finding.