Back to skill

Security audit

llm-gateway-hub

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent local LLM gateway skill, but users should handle API keys and any synced gateway files carefully.

Use this only if you intend to store model-provider API keys locally and send prompts to the configured providers. Keep gateway_config.json out of cloud sync, repos, shared folders, and chat logs; restrict file permissions manually; verify provider base URLs and pricing before use; and review the image watermark setting if provenance marking matters in your environment.

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
  • 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 (20)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The skill metadata and documentation strongly advertise a unified LLM gateway with key management, budget enforcement, and spend tracking, but the finding indicates the actual implementation does not provide those controls and instead adds unrelated security-report visualization behavior. This kind of description-behavior mismatch is dangerous because operators may trust the skill with API integration, secrets, or governance decisions under false assumptions, leading to silent control gaps and misuse in production.

Content

No source excerpt is available for this finding.

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 100)May include surrounding context.

md
| 文档与理论文本 | 本 `SKILL.md`、`README.md`、其中「理论依据」段与一切理论表述 | **不在 MIT 覆盖范围内**:保留所有权利(All rights reserved) |

Ae1

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 117)May include surrounding context.

md
| 文档与理论文本 | 本 `SKILL.md`、`README.md`、其中「理论依据」段与一切理论表述 | **不在 MIT 覆盖范围内**:保留所有权利(All rights reserved) |

Lp1

High
Category
MCP Least Privilege
Confidence
75% confidence
Finding

The skill uses 'env' capability that is not listed in its permissions. This may indicate deceptive intent or missing permission declarations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The README explicitly promotes having a company computer read the local gateway state via cloud file synchronization, but it does not warn about the security implications of synchronizing sensitive gateway data across third-party cloud storage. In the context of an LLM gateway that manages API keys, budget ledgers, routing state, and potentially prompts or usage logs, this can expand the attack surface and lead to credential leakage, metadata exposure, or accidental cross-device data disclosure.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

Overly broad trigger conditions can cause the skill to activate in contexts beyond its intended scope, increasing the chance of unintended tool use, user confusion, or accidental interception of queries unrelated to this skill. In a skill claiming to manage model gateways, keys, and costs, broad invocation is more risky because users may expose operational or secret-bearing context to the wrong capability surface.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The natural-language note is written as a mandatory Chinese-only instruction set and does not provide any language or locale option. Under the policy, forcing a specific language without user opt-in is a reportable natural-language policy violation unless the locale restriction is explicitly justified, which is not stated here.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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/gateway_config.template.json (reported line 31)May include surrounding context.

json
},
    "deepseek": {
      "enabled": false,
      "base_url": "https://api.deepseek.com/v1",
      "api_key": "<在此填入你的_API_Key>",
      "default_text_model": "deepseek-chat",
      "default_image_model": "",

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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/gateway_config.template.json (reported line 42)May include surrounding context.

json
},
    "kimi": {
      "enabled": false,
      "base_url": "https://api.moonshot.cn/v1",
      "api_key": "<在此填入你的_API_Key>",
      "default_text_model": "moonshot-v1-8k",
      "default_image_model": "",

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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/gateway_config.template.json (reported line 64)May include surrounding context.

json
},
    "hunyuan": {
      "enabled": false,
      "base_url": "https://api.hunyuan.cloud.tencent.com/v1",
      "api_key": "<在此填入你的_API_Key>",
      "default_text_model": "hunyuan-turbo",
      "default_image_model": "",

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% 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/gateway_config.template.json (reported line 75)May include surrounding context.

json
},
    "siliconflow": {
      "enabled": false,
      "base_url": "https://api.siliconflow.cn/v1",
      "api_key": "<在此填入你的_API_Key>",
      "default_text_model": "deepseek-ai/DeepSeek-V3",
      "default_image_model": "black-forest-labs/FLUX.1-schnell",

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The image generation request hard-codes watermark to False, causing generated images to be produced without provenance marking or user consent. In a shared enterprise gateway, this increases misuse risk, weakens traceability, and may violate provider policy or internal governance requirements for AI-generated media.

Content

No source excerpt is available for this finding.

Dynamic import via __import__()

Medium
Category
Dangerous Code Execution
Confidence
75% confidence
Finding

Dynamic import() can load arbitrary modules at runtime, bypassing static analysis and potentially importing malicious code.

Content

Scanner excerpt · scripts/gen_security_radar.py (reported line 27)May include surrounding context.

python
ang = -90 + 360.0 / n * i
        rad = ang * 3.14159265358979 / 180.0
        rr = r * (v / max_v)
        x = cx + rr * pow(abs(__import__("math").cos(rad)), 1) * (1 if __import__("math").cos(rad) >= 0 else -1)
        y = cy + rr * pow(abs(__import__("math").sin(rad)), 1) * (1 if __import__("math").sin(rad) >= 0 else -1)
        pts.append((x, y))
    return pts

Dynamic import via __import__()

Medium
Category
Dangerous Code Execution
Confidence
75% confidence
Finding

Dynamic import() can load arbitrary modules at runtime, bypassing static analysis and potentially importing malicious code.

Content

Scanner excerpt · scripts/gen_security_radar.py (reported line 28)May include surrounding context.

python
rad = ang * 3.14159265358979 / 180.0
        rr = r * (v / max_v)
        x = cx + rr * pow(abs(__import__("math").cos(rad)), 1) * (1 if __import__("math").cos(rad) >= 0 else -1)
        y = cy + rr * pow(abs(__import__("math").sin(rad)), 1) * (1 if __import__("math").sin(rad) >= 0 else -1)
        pts.append((x, y))
    return pts

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The file begins with a Chinese-language scope notice and does not provide an equivalent user-selectable language option for the policy-bearing explanation. Under the stated policy, forcing a specific language without opt-in can be a natural-language policy violation unless the locale constraint is justified, which is not indicated here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The module docstring, CLI descriptions, status messages, and errors are all written in Chinese, and the skill does not offer locale selection or user opt-in for language. This can violate language/locale policy when tools are expected to adapt to user preference rather than forcing a single language.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The module docstring and generated output are written entirely in Chinese, and the HTML is explicitly emitted with lang="zh-CN" later in the file. This indicates a fixed language/locale behavior without any visible opt-in, selection mechanism, or justification that the skill is region-specific.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The document sets lang="zh-CN" and all visible UI text is presented only in Simplified Chinese. For a general skill artifact, this imposes a specific language/locale without indicating that the user can choose another language or that the file is intentionally region-specific.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

This JSON file uses Chinese-only natural-language strings for the display name, mode, dimension names, and check details. Because no opt-in, alternate locale, or documented region-specific scope is provided in the file, it may violate the policy against forcing a specific language without user choice.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
80% confidence
Finding

该文件整体仅以中文呈现审计说明、结论和使用建议,未见提供语言/地区选择、双语支持,或说明该技能/文档仅面向特定中文环境。根据规则,若技能或文档强制特定语言而无用户选择或明确合理限定,属于自然语言层面的语言/locale 策略风险。

Content

No source excerpt is available for this finding.