Back to skill

Security audit

捷帮图片工具

Security checks for vulnerabilities and agentic risk

Overview

The skill does what it claims, but it sends user images and QR-code text to an external service with an admin-style credential and does not clearly disclose those privacy and privilege implications.

Review before installing. Avoid using this skill with private photos, confidential screenshots/documents, internal image URLs, tokens, password-reset links, payment data, or other secrets in QR codes unless you trust the jiebang.site service and its logging/retention practices. The publisher should disclose remote processing, avoid GET for QR payloads, and replace the admin-style API key with a narrowly scoped service credential.

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)

other

Warning
Location
scripts/main.py:44
Finding

User-Supplied Image Data Is Transmitted to an Undisclosed External Service

Content
View full analysis

Vulnerability Details

File Location: scripts/main.py, lines 44–55
Vulnerability Type: Undisclosed Data Transmission
Risk Level: Medium

Vulnerable Code

python
api_url = f"{BASE_URL}/api/image-convert"

payload = {
    "image": image,
    "from": from_format,
    "to": to_format,
    "quality": quality
}

try:
    response = requests.post(
        api_url,
        headers=get_headers(),
        json=payload,
        timeout=60
    )

BASE_URL is defined on line 13 as:

python
BASE_URL = "https://www.jiebang.site"

Technical Analysis

The image-conversion function submits the complete user-provided image value to https://www.jiebang.site/api/image-convert. This value may contain an image URL or the complete image encoded as Base64.

Remote processing is not explicitly disclosed in SKILL.md; its functional description may consequently lead users to believe that conversion is performed locally. There is also no code-level control for obtaining user consent, minimizing transmitted data, or communicating the external service's retention and privacy policies.

This behavior does not provide the external service with local system privileges. Its security impact is limited primarily to the confidentiality and handling of the content deliberately submitted for conversion.

Attack Path

  1. A user asks the skill to convert an image containing private or confidential information.
  2. The image or its Base64 representation is passed to image_convert.
  3. The function includes the complete value in a JSON payload.
  4. The payload is transmitted to the external www.jiebang.site service.
  5. The external service and its supporting infrastructure can process, log, or retain the submitted content according to controls that are not documented in the project.

Impact Assessment

The affected scope is the confidentiality of images submitted to the conver ...[truncated 439 chars]

Remediation
View remediation

Remediation Suggestions

  1. Clearly disclose in SKILL.md that images and image URLs are transmitted to https://www.jiebang.site for processing.
  2. Obtain explicit user consent before transmitting potentially sensitive content.
  3. Document the service operator, retention period, deletion process, geographic processing location, and applicable privacy policy.
  4. Prefer local conversion through a vetted image-processing library when remote processing is unnecessary.
  5. Minimize submitted data and remove metadata where it is not required for conversion.
  6. Apply contractual and technical controls that prevent server-side request logging or long-term storage of submitted images.
  7. Use a narrowly scoped API credential and avoid sharing one administrative credential across unrelated users or workloads.

T09 · Insecure Skill Coding Practices

Warning
Location
scripts/main.py:78
Finding

Sensitive QR-Code Content Is Placed in HTTP Query Parameters

Content
View full analysis

Vulnerability Details

File Location: scripts/main.py, lines 78–90
Vulnerability Type: Sensitive Data Exposure Through URL Query Parameters
Risk Level: Medium

Vulnerable Code

python
api_url = f"{BASE_URL}/api/qrcode"

try:
    # 尝试GET请求
    params = {"text": text, "size": size, "error_level": error_level}
    response = requests.get(
        api_url,
        headers=get_headers(),
        params=params,
        timeout=30
    )
    
    if response.status_code >= 400:
        # 尝试POST请求

Technical Analysis

The QR-code generator initially issues an HTTP GET request and places the complete user-provided text value in the URL query string. Although HTTPS protects the request in transit, query strings are commonly captured by origin-server access logs, reverse proxies, gateways, application monitoring systems, and diagnostic tooling.

QR codes can contain confidential material such as authentication links, password-reset URLs, enrollment secrets, internal addresses, personal data, or payment information. Placing that material in a URL unnecessarily expands the number of systems where it may persist.

Falling back to POST after an error does not mitigate the issue because the initial GET request, including its query string, has already been sent.

Attack Path

  1. A user requests a QR code containing sensitive text or a secret-bearing URL.
  2. generate_qrcode copies that content into the text query parameter.
  3. The HTTP client constructs a URL containing the complete QR payload.
  4. The request passes through external service infrastructure.
  5. The full URL may be recorded in access logs, reverse-proxy logs, observability platforms, or error traces.
  6. A person or service with access to those records can recover the QR payload and potentially use any embedded secret.

Impact Assessment

The issue can expose only the QR content supplied by the user; it does ...[truncated 505 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove the GET request and use POST exclusively for QR-code generation.
  2. Put text, size, and error_level in a JSON request body rather than the URL.
  3. Configure the origin server, proxies, and observability systems to redact QR payloads and other sensitive request fields.
  4. Establish short log-retention periods and restrict access to request logs.
  5. Avoid including response bodies containing potentially sensitive details in raised exceptions.
  6. Warn users not to encode long-lived credentials or reusable secrets in QR codes.
  7. Prefer local QR generation with a vetted library when external processing is not required.
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
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (6)

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
90% confidence
Finding

The skill advertises capabilities that imply network and possibly environment access, but it does not declare any explicit tool scope or permission boundaries. This can lead to overbroad execution in host environments, making it harder to constrain outbound requests or access to sensitive runtime context if the implementation behind the skill is invoked.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The manifest description says to use the skill when the user mentions phrases like '图片格式转换' and especially '二维码生成', which are generic request patterns rather than narrowly scoped activation rules. It does not provide exclusion conditions or negative examples, so the trigger boundary is ambiguous and may cause unintended invocation for ordinary conversation about images or QR codes.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill accepts public image URLs and embeds user-provided content into generated QR codes or output images, but it does not warn users about privacy, tracking, or unintended disclosure. Users may unknowingly expose private images to remote servers, include sensitive data in QR codes, or generate artifacts that leak information when shared.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill retrieves a privileged admin credential from the environment and sends it as an X-Admin-Key for routine image conversion and QR generation operations. Using admin-level authentication for ordinary user-triggered actions violates least privilege and makes any misuse of the skill, downstream API abuse, or accidental credential exposure significantly more damaging.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The image conversion function transmits user-supplied image content or image URLs to a third-party remote service without any in-code disclosure, consent flow, or data handling notice. This is risky because images may contain sensitive content or metadata, and URL submission can also cause the remote service to fetch internal or private resources on behalf of the user workflow.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The QR generation function sends arbitrary user-provided text to a remote API via GET and fallback POST without a clear user-facing warning. QR payloads can contain sensitive links, tokens, personal data, or internal identifiers, and use of GET additionally risks exposure through logs, caches, and intermediaries.

Content

No source excerpt is available for this finding.