Back to skill

Security audit

arkcli-doctor

Security checks for vulnerabilities and agentic risk

Overview

The verdict is benign with medium confidence: the scan reports a skill focused on Ark CLI diagnostics and media-origin checks, with findings about scope mismatches and privacy disclosure, while some reference checks were incomplete.

The verdict is benign, not a guarantee of zero risk. Before installing, consider whether the reported scope mismatches, elevated clock-repair commands, and missing disclosure about sending media URLs to an external verification service fit your needs. Skillspector’s analysis is partial: it inspected all 11 listed components but could not resolve two local references, so the evidence is incomplete. The AIG scan completed with no findings.

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
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (37)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied shell commands create new media rather than diagnose CLI or service issues or verify the origin of existing media. Although the file is named error-codes.md, its commands implement a media-generation workflow, a materially different capability not declared in the doctor skill description.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied command performs generative inference, which is not among the declared diagnostic or origin-verification capabilities. Although it appears in an error-code reference and could be suggested remediation, it initiates content generation rather than inspecting or diagnosing an existing issue, making it a materially distinct capability.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

Although CLI installation could support later diagnostics, the declared scope expressly excludes installation. The supplied chunk solely performs that excluded operation, so its behavior does not match the stated scope.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

该片段虽然位于错误码参考文档中,但实际给出的是媒体重新生成命令,而不是错误分析或来源验证命令。生成操作会使用模型及参考图创建新内容,属于声明中未涵盖的能力;它也不是完成 doctor 诊断或来源验证所必需的实现细节。注释仅指导裁剪人脸,片段本身未执行裁剪。

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
93% confidence
Finding

The verification operation, batch size, and JSON format match the declared purpose. However, --yes bypasses the confirmation gate. The supplied chunk contains no prior fee disclosure or user-confirmation check, so it does not implement the explicitly declared requirement to obtain confirmation before issuing verification business requests.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
92% confidence
Finding

The supplied command initiates generation using the original model and prompt. Although it appears in an error-code reference and might be offered as remediation, the command itself performs a generation operation not declared in the diagnostic and origin-verification purpose. Disabling audio does not make that operation diagnostic.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
88% confidence
Finding

Although suggesting safer inputs could support content-moderation troubleshooting, this command performs generation itself. The declared purpose covers diagnosing moderation errors but does not declare content generation. The mismatch is the generation operation, not the prompt revision or use of replacement material.

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 · SKILL.md (reported line 25)May include surrounding context.

md
本 skill 一次性覆盖八个平行 domain。每个 domain 一份 reference,按用户场景路由:

| Domain                | 命令                                | Reference                                                                                       | 解决什么                                                                  |
|-----------------------|-------------------------------------|-------------------------------------------------------------------------------------------------|---------------------------------------------------------------------------|
| **CLI 健康**(默认)  | `arkcli doctor`                     | [`references/scope-cli.md`](references/scope-cli.md)                                            | 本地二进制版本 / 网络联通 / 时钟偏差 / 当前 profile 认证——onboarding 首查 |
| **account**           | `arkcli doctor account`             | [`references/scope-account.md`](references/scope-account.md)                                    | identity / realname / balance / IAM Ark 系统预置策略 / VMP 三段 / TOS 开通  |

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 · references/error-codes.md (reported line 43)May include surrounding context.

md
Ark 数据面调用时返回的错误码——内容安全 / 版权 / 隐私拦截、风险识别、参数 / 格式校验、鉴权 / 模型权限、限流、账号 / 计费状态、资源状态、精调操作约束。

| HTTP | Type            | 错误码                                                          | 释义                                                       | 处理方式                                                                                       |
|------|-----------------|----------------------------------------------------------------|------------------------------------------------------------|------------------------------------------------------------------------------------------------|
| 400  | BadRequest      | `MissingParameter`                                             | 缺必参                                                      | 按 message 里的字段名 + API 文档补                                                            |
| 400  | BadRequest      | `MissingParameter.{{Parameter}}`                               | 缺指定的请求参数                                            | 按 message 提示的字段名补

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 · references/scope-report.md (reported line 15)May include surrounding context.

md
`arkcli doctor report` 就是这一步的执行入口。

| 用户场景                                                        | Agent 跑什么                                                                                       |
|----------------------------------------------------------------|----------------------------------------------------------------------------------------------------|
| 「我这次 seedance 2.x 生成的视频角色脸变形了 / 字幕拼错了」     | **未明确说上报,不直达 report**;有模型名走 `doctor model`,有 ep-id 走 `doctor infer-endpoint`        |
| 「先检查这次 seedance 2.x 字幕错的原因,再帮我上报」           | **诊断意图优先**;先走 `doctor model` / `doctor infer-endpoint`,完成后按 Path A 询问并上报             |

The origin-verification workflow lacks a privacy warning that user-provided media URLs are submitted to an external service.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The workflow instructs the agent to submit all supplied media URLs and create remote verification tasks after obtaining confirmation of fees. Although remote execution and possible charges are disclosed elsewhere, the skill does not warn that URLs—including any embedded access tokens—and the referenced media may become accessible to the verification provider.

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 · SKILL.md (reported line 158)May include surrounding context.

md
### 完整 subtype 路由

| 业务领域       | subtype                  | 跳节                                                                                                  |
|----------------|--------------------------|------------------------------------------------------------------------------------------------------|
| 方舟错误码     | `input_real_face`        | [`error-codes.md` §1.1.1](references/error-codes.md#111-input_real_face)                              |
| 方舟错误码     | `input_copyright`        | [`error-codes.md` §1.1.2](references/error-codes.md#112-input_copyright)                              |

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 · SKILL.md (reported line 198)May include surrounding context.

md
这条命令是**只读查表**——不需登录、不消耗推理 token——把一个错误码翻译成结构化诊断。返回字段:

| 字段              | 含义                                                                                            |
|-------------------|------------------------------------------------------------------------------------------------|
| `category`        | 顶层类别(如 `policy` / `permission` / `quota` / `infrastructure`)                              |
| `subtype`         | 类别下的子类型(如 `input_real_face` / `model_access_denied` / `rate_limit_exceeded`)            |

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 · references/error-codes.md (reported line 59)May include surrounding context.

md
超 context                          | 减少图 / 截断文本 / 换长 context 模型                                                          |
| 400  | BadRequest      | `InvalidEndpoint.ClosedEndpoint`                               | 推理接入点已关闭或暂时不可用                                | 联系管理员或重启:`arkcli infer endpoint start`                                                |
| 400  | BadRequest      | `Duplicate.Tags.Key`                                           | tag 列表中存在重复 key                                      | 去重                                                                                           |
| 400  | BadRequest      | `InputTextSensitiveContentDetected`                            | 输入文本敏感                                                | 改写 prompt(生视频场景见 [§1.1.3](#113-input_content_safety))                              |
| 400  | BadRequest      | `InputImageSensitiveContentDetected`                           | 输入图敏感                                                  | 换素材(生视频场景见 [§1.1.3](#113-

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 · references/error-codes.md (reported line 69)May include surrounding context.

md
| 见 [§1.1.1](#111-input_real_face)                                                              |
| 400  | BadRequest      | `Input{Text,Image,Video,Audio}SensitiveContentDetected.PolicyViolation` | 输入素材涉版权                                       | 见 [§1.1.2](#112-input_copyright)                                                              |
| 400  | BadRequest      | `InputTextRiskDetection`                                       | 风险识别产品 - 输入文本拦截                                  | 改写 prompt                                                                                    |
| 400  | BadRequest      | `InputImageRiskDetection`                                      | 风险识别产品 - 输入图拦截                                    | 换素材                                                                                          |
| 400  | BadRequest      | `OutputTextRiskDetection`                                      | 风险识别产品 - 输出文本拦截                                  | 收敛 prompt

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 · references/error-codes.md (reported line 71)May include surrounding context.

md
| 改写 prompt                                                                                    |
| 400  | BadRequest      | `InputImageRiskDetection`                                      | 风险识别产品 - 输入图拦截                                    | 换素材                                                                                          |
| 400  | BadRequest      | `OutputTextRiskDetection`                                      | 风险识别产品 - 输出文本拦截                                  | 收敛 prompt                                                                                    |
| 400  | BadRequest      | `OutputImageRiskDetection`                                     | 风险识别产品 - 输出图拦截                                    | 收敛 prompt / 换风格                                                                           |
| 400  | BadRequest      | `ContentSecurityDetectionError`                                | 风险识别产品**本身**调用失败                                | 重试;持续失败提工单(带 CSDRequestId)

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 · references/error-codes.md (reported line 74)May include surrounding context.

md
| 收敛 prompt / 换风格                                                                           |
| 400  | BadRequest      | `ContentSecurityDetectionError`                                | 风险识别产品**本身**调用失败                                | 重试;持续失败提工单(带 CSDRequestId)                                                       |
| 400  | BadRequest      | `SensitiveContentDetected`                                     | 输入文本敏感(旧版兼容)                                    | 改写 prompt                                                                                    |
| 400  | BadRequest      | `SensitiveContentDetected.SevereViolation`                     | 输入文本严重违规                                            | 改写 prompt                                                                                    |
| 400  | BadRequest      | `SensitiveContentDetected.Violence`                            | 输入文本含激进行为                                          | 改写 prompt

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 · references/error-codes.md (reported line 75)May include surrounding context.

md
| 重试;持续失败提工单(带 CSDRequestId)                                                       |
| 400  | BadRequest      | `SensitiveContentDetected`                                     | 输入文本敏感(旧版兼容)                                    | 改写 prompt                                                                                    |
| 400  | BadRequest      | `SensitiveContentDetected.SevereViolation`                     | 输入文本严重违规                                            | 改写 prompt                                                                                    |
| 400  | BadRequest      | `SensitiveContentDetected.Violence`                            | 输入文本含激进行为                                          | 改写 prompt                                                                                    |
| 400  | Forbidden       | `InvalidSubscription`                                          | Coding Plan 套餐未订阅或已过期                              | 续订(走 `arkcli-plans`,本仓未独立成 sk

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 · references/error-codes.md (reported line 76)May include surrounding context.

md
| 改写 prompt                                                                                    |
| 400  | BadRequest      | `SensitiveContentDetected.SevereViolation`                     | 输入文本严重违规                                            | 改写 prompt                                                                                    |
| 400  | BadRequest      | `SensitiveContentDetected.Violence`                            | 输入文本含激进行为                                          | 改写 prompt                                                                                    |
| 400  | Forbidden       | `InvalidSubscription`                                          | Coding Plan 套餐未订阅或已过期                              | 续订(走 `arkcli-plans`,本仓未独立成 skill 时回 [`arkcli-shared`](../../arkcli-shared/SKILL.md)) |
| 401  | Unauthorized    | `AuthenticationError`                                          | API Key / AK-SK 校验未通过                                  | 重新登录:`arkcli auth login`(走 [`arkc

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 · references/error-codes.md (reported line 95)May include surrounding context.

md
| 见 [§1.2](#12-modelaccessdenied) 修复 A                                                       |
| 404  | NotFound        | `NotFound.{{Parameter}}`                                       | 指定资源找不到                                              | 按 message 提示的参数名核对                                                                   |
| 404  | NotFound        | `UnsupportedModel`                                             | 模型不支持 Coding Plan                                      | 换支持的模型                                                                                    |
| 429  | TooManyRequests | `RateLimitExceeded.EndpointRPMExceeded`                        | endpoint RPM 撞顶                                            | 见 [§1.3](#13-ratelimitexceeded)(接入点级瓶颈)                                              |
| 429  | TooManyRequests | `RateLimitExceeded.EndpointTPMExceeded`                        | endpoint TPM 撞顶                                            | 见 [§1.3](#13-ratelimi

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 · references/error-codes.md (reported line 70)May include surrounding context.

md
| 见 [§1.1.2](#112-input_copyright)                                                              |
| 400  | BadRequest      | `InputTextRiskDetection`                                       | 风险识别产品 - 输入文本拦截                                  | 改写 prompt                                                                                    |
| 400  | BadRequest      | `InputImageRiskDetection`                                      | 风险识别产品 - 输入图拦截                                    | 换素材                                                                                          |
| 400  | BadRequest      | `OutputTextRiskDetection`                                      | 风险识别产品 - 输出文本拦截                                  | 收敛 prompt                                                                                    |
| 400  | BadRequest      | `OutputImageRiskDetection`                                     | 风险识别产品 - 输出图拦截                                    | 收敛 prompt / 换风格

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 · references/scope-infer-endpoint.md (reported line 75)May include surrounding context.

md
**错误码路由**——按 doctor 总入口的 references 表:

| 错误码                                    | 加载 reference                                                                                          |
|-------------------------------------------|---------------------------------------------------------------------------------------------------------|
| `ModelAccessDenied`                       | [`error-codes.md` §1.2](error-codes.md#12-modelaccessdenied)                                            |
| `RateLimitExceeded` / 429                 | [`error-codes.md` §1.3](error-codes.md#13-ratelimitexceeded)                                            |

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 · references/error-codes.md (reported line 88)May include surrounding context.

md
关联资源处于非空闲态(Context InProgress / 缓存更新中 / File 非可用) | 等待任务完成 / 取消 / 换 ID                                                                   |
| 403  | Forbidden       | `OperationDenied.UnsupportedPhase`                             | 资源处于特殊状态(锁定 / 处理中等)                         | 检查资源状态;不可解则提工单                                                                  |
| 403  | Forbidden       | `OperationDenied.ConflictedValidationSet`                      | 同时上传验证集 + 设置训练集采样为验证集百分比               | 二选一(精调)                                                                                 |
| 403  | Forbidden       | `OperationDenied.UnsupportedCustomizationType`                 | 模型不支持该精调方法                                         | 换方法或换模型                                                                                 |
| 403  | Forbidden       | `OperationDenied.CustomizationNotSupported`                    | 基础模型版本不支持该精调方法                                 | 换基础模型版本

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 · references/error-codes.md (reported line 89)May include surrounding context.

md
于特殊状态(锁定 / 处理中等)                         | 检查资源状态;不可解则提工单                                                                  |
| 403  | Forbidden       | `OperationDenied.ConflictedValidationSet`                      | 同时上传验证集 + 设置训练集采样为验证集百分比               | 二选一(精调)                                                                                 |
| 403  | Forbidden       | `OperationDenied.UnsupportedCustomizationType`                 | 模型不支持该精调方法                                         | 换方法或换模型                                                                                 |
| 403  | Forbidden       | `OperationDenied.CustomizationNotSupported`                    | 基础模型版本不支持该精调方法                                 | 换基础模型版本                                                                                 |
| 404  | NotFound        | `InvalidEndpointOrModel.NotFound`                              | 模型 / endpoint 不存在或无权访问                             | 检查 ID + profile region;可能是 [§1.2](#12-mod

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 · references/error-codes.md (reported line 90)May include surrounding context.

md
集采样为验证集百分比               | 二选一(精调)                                                                                 |
| 403  | Forbidden       | `OperationDenied.UnsupportedCustomizationType`                 | 模型不支持该精调方法                                         | 换方法或换模型                                                                                 |
| 403  | Forbidden       | `OperationDenied.CustomizationNotSupported`                    | 基础模型版本不支持该精调方法                                 | 换基础模型版本                                                                                 |
| 404  | NotFound        | `InvalidEndpointOrModel.NotFound`                              | 模型 / endpoint 不存在或无权访问                             | 检查 ID + profile region;可能是 [§1.2](#12-modelaccessdenied)                                |
| 404  | NotFound        | `InvalidEndpointOrModel.ModelIDAccessDisabled`                 | 不允许直接用 model ID 调用,需用 endpoint ID                 | 用 endpoint ID 调(不要直接 model ID)