Back to skill

Security audit

蛋糕叔叔商城点单助手

Security checks for vulnerabilities and agentic risk

Overview

This shopping skill appears purpose-aligned, but it uses broad command execution and stores account and delivery data in plaintext, so it needs review before installation.

Review this skill before installing. It is not clearly malicious, but it can handle account tokens, saved addresses, phone numbers, and order links while using a broad local command tool. Only use it if you trust the publisher and service, understand that a plaintext local token file will be created, and are comfortable with the agent making account-linked shopping API calls from your machine.

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 (2)

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:18
Finding

Shell Command Injection Through User-Controlled JSON Arguments

Content
View full analysis
''" ``` ```bash python "~/.workbuddy/skills/dgss/scripts/dgss.py" '' ``` The same command-construction pattern is documented for operations that process user-controlled product keywords, names, phone numbers, and delivery addresses: ```bash python "~/.workbuddy/skills/dgss/scripts/dgss.py" addAddrList '{"user_token":"TOKEN","province":"Beijing","city":"Beijing","area":"Haidian","addr":"DETAILED_ADDRESS","name":"NAME","phone":"PHONE"}' ``` The tool configuration exposes the entire invocation as a free-form command string: ```json { "properties": { "command": { "type": "string" }, "explanation": { "type": "string" }, "requires_approval": { "type": "boolean" } }, "required": ["command", "explanation"] } ``` ### Technical Analysis The Skill directs the Agent to interpolate JSON containing user-controlled values into a shell command. On Linux and macOS, the JSON is enclosed in single quotes. A malicious value containing a single quote can terminate that quoted argument and introduce shell metacharacters such as `;`, `|`, command substitution, or redirection. Windows commands are similarly assembled as strings passed through `cmd /c`, creating platform-specific injection opportunities involving quote termination and command separators. The Python script only removes matching quotes and parses the resulting argument: ```python raw_args = sys.argv[2] if raw_args.startswith("'") and raw_args.endswith("'"): raw_args = raw_args[1:-1] elif raw_args.startswith('"') and raw_args.endswith('"'): raw_args = raw_args[1:-1] args = json.loads(raw_args) ` ...[truncated 1938 chars]
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/dgss.py:91
Finding

Account Token Persisted in Plaintext and Exposed Through Process Arguments

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (34)

Tp2

High
Category
MCP Tool Poisoning
Confidence
85% confidence
Finding

Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The skill claims it operates purely via curl and does not rely on an MCP proxy, yet the documentation directs execution through a Python script and additional capabilities such as order-status lookup. This kind of description-behavior mismatch weakens user trust and reviewability, making it easier to hide data handling or network behavior that users and auditors did not consent to.

Content

No source excerpt is available for this finding.

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill claims it operates purely via curl and does not rely on an MCP proxy, yet the documentation directs execution through a Python script and additional capabilities such as order-status lookup. This kind of description-behavior mismatch weakens user trust and reviewability, making it easier to hide data handling or network behavior that users and auditors did not consent to.

Content

No source excerpt is available for this finding.

Ssd 3

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The defined markdown profile stores user_token together with default city, address ID, and last order information in one place. In the context of a commerce skill, this is a high-risk data concentration pattern because compromise of a single local file can expose credentials, personal data, and transaction context sufficient for misuse or social engineering.

Content

No source excerpt is available for this finding.

Lp1

High
Category
MCP Least Privilege
Confidence
94% confidence
Finding

The file invokes shell functionality via os.system without declaring shell capability in permissions metadata. Even though the current command is fixed, undeclared shell access weakens platform security expectations and can enable unsafe expansion later or bypass operator review of privileged behavior.

Content

No source excerpt is available for this finding.

Lp1

High
Category
MCP Least Privilege
Confidence
94% confidence
Finding

The file invokes shell functionality via os.system without declaring shell capability in permissions metadata. Even though the current command is fixed, undeclared shell access weakens platform security expectations and can enable unsafe expansion later or bypass operator review of privileged behavior.

Content

No source excerpt is available for this finding.

os.system() or os exec-family call

High
Category
Dangerous Code Execution
Confidence
85% confidence
Finding

os.system() and os exec-family calls run shell commands with the process's full privileges, enabling arbitrary command execution.

Content

Scanner excerpt · scripts/dgss.py (reported line 13)May include surrounding context.

python
# 设置控制台编码为 UTF-8
if sys.platform == 'win32':
    os.system('chcp 65001 > nul')
    sys.stdout.reconfigure(encoding='utf-8')

def call_dangaoss(tool_name, arguments):

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The only exposed tool is a generic execute_command interface that accepts an arbitrary string command, which grants shell execution far beyond the stated shopping-assistant purpose. In a shopping skill context, this is especially risky because natural-language requests could be transformed into unrestricted OS commands, enabling file access, data exfiltration, or execution of unintended programs if the agent or downstream components are manipulated.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The manifest advertises a narrower capability set than the body of the skill actually implements, including order-status queries and reuse of stored order data for repeat purchases. Undisclosed capabilities matter in agent skills because they expand the operational surface and may cause users or reviewers to miss privacy and authorization implications.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The trigger conditions include broad shopping-related keywords such as '蛋糕', '下单', and '订购', which can cause the skill to activate in contexts beyond the intended domain. Over-broad activation is risky for a skill that reads files, stores personal data, and can initiate ordering flows because it increases the chance of unintended handling of sensitive user input.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill instructs storing a user token in a local markdown file, which is an unsafe persistence mechanism for credentials and adjacent personal data. Local plaintext storage increases the chance of accidental disclosure through backups, sync tools, filesystem access, or later model/tool reads, enabling account misuse if the token is recovered.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill tells the agent to persist the user's token locally without a clear user-facing warning or opt-in around credential retention. In a shopping/account context, that creates meaningful privacy and account-takeover risk because a recovered token may authorize address access, order history access, or purchase actions.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The instructions require persistent storage and reuse of token, address, phone, and order data in local markdown files. This is dangerous because these files are easy to read, copy, sync, or accidentally expose, and the combined dataset enables both privacy leakage and unauthorized account-linked actions.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill explicitly reads back stored address context and offers reuse of prior address data in later interactions. In this context, that increases the chance of exposing personal delivery information to the wrong conversation, user, or tool invocation if session boundaries or file access are not tightly enforced.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

Repeatedly saving address identifiers after interactions expands the amount of retained personal state and normalizes long-term tracking of user delivery preferences. While an address ID alone may seem limited, in combination with token and city it facilitates profiling and account-linked actions.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill instructs the agent to display recipient name, phone number, and full delivery address in the order summary. Exposing full PII in conversational output increases shoulder-surfing, transcript leakage, and unintended log retention risks, especially in shared environments.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

Persistently updating order history fields adjacent to the user token creates a durable linkage between credential material and purchase behavior. This increases privacy risk and can help an attacker or another tool correlate identity, account access, and order metadata from a single plaintext file.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The file requires that all preset copy be used exactly as written and those prompts are in Chinese, effectively forcing a specific language for user interactions. There is no user opt-in or documented justification that this skill must operate only in Chinese, so this creates a language policy concern.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill defines persistent storage for address IDs, city, and order history fields tied to a user token without a clear privacy warning or retention policy. This accumulates personal data over time and increases the blast radius if the local memory files are exposed or reused by other tools.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · references/api_docs.md (reported line 27)May include surrounding context.

| user_token | string | 是 | 用户 token | | city_name | string | 是 | 城市名(规范到市一级,如"深圳市") |

curl 示例:

bash
curl -s -X POST "https://www.dangaoss.com/dsapi/workbuddy/mcp_server" \
  -H "Content-Type: application/json" \

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · references/api_docs.md (reported line 88)May include surrounding context.

| user_token | string | 是 | 用户 token | | city_name | string | 是 | 城市名(规范到市一级,如"深圳市") |

curl 示例:

bash
curl -s -X POST "https://www.dangaoss.com/dsapi/workbuddy/mcp_server" \
  -H "Content-Type: application/json" \

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · references/api_docs.md (reported line 117)May include surrounding context.

| user_token | string | 是 | 用户 token | | city_name | string | 是 | 城市名(规范到市一级,如"深圳市") |

curl 示例:

bash
curl -s -X POST "https://www.dangaoss.com/dsapi/workbuddy/mcp_server" \
  -H "Content-Type: application/json" \

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · references/api_docs.md (reported line 144)May include surrounding context.

| user_token | string | 是 | 用户 token | | city_name | string | 是 | 城市名(规范到市一级,如"深圳市") |

curl 示例:

bash
curl -s -X POST "https://www.dangaoss.com/dsapi/workbuddy/mcp_server" \
  -H "Content-Type: application/json" \

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The document instructs the agent to transmit sensitive personal data including recipient name, phone number, and detailed address to an external service, but provides no privacy warning, data minimization guidance, retention notice, or user-consent requirement. In a shopping/order skill this creates a real privacy and compliance risk because the skill operationally handles PII and could encourage collection or transmission without adequate safeguards or user awareness.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
91% confidence
Finding

This example transmits high-sensitivity personal data—recipient name, phone number, and full delivery address—plus authentication material to an external service. In the context of an agent skill, that is dangerous if consent, secure handling, redaction, and logging restrictions are not explicitly enforced, because it increases the chance of unauthorized disclosure of PII or misuse of account-linked order data.

Content

Scanner excerpt · references/api_docs.md (reported line 59)May include surrounding context.

| name | string | 是 | 收货人姓名 | | phone | string | 是 | 11位手机号(1[3-9]开头) |

curl 示例:

bash
curl -s -X POST "https://www.dangaoss.com/dsapi/workbuddy/mcp_server" \
  -H "Content-Type: application/json" \