Back to skill

Security audit

优稿降 AI 改写与复检

Security checks for vulnerabilities and agentic risk

Overview

The skill matches its stated purpose and provides clear controls for paid text rewriting, detection, and retries.

Use a key limited to rewriting and detection, submit only text you are comfortable sharing with Yougao, and set a points cap because none is enabled by default. Shared helpers also support publishing, which requires separate authorization and permissions. Protect saved credentials and manuscript files; disconnecting locally does not revoke the key. This review does not verify the live service's retention practices.

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
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • System Prompt LeakageDirect Leakage, Indirect Extraction, Tool-Based Exfiltration
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
Findings (23)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The supplied chunk only implements a table-of-contents plugin. It contains no 优稿 integration, AI-content detection, rewriting, rechecking, target-based retry logic, points accounting, or version retention. Its vendor-library location suggests it could support a larger application, but this chunk provides no visible connection between its Markdown-rendering functionality and the declared workflow. This finding is limited to the supplied chunk; no external access or malicious behavior is evident.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The chunk contains no 优稿 integration, AI-score detection, rewriting, rechecking, target-based retries, points-budget enforcement, or original/best-version retention. Instead, its executable entry point implements a Markdown converter. Although its vendor path suggests it may be a supporting dependency, the supplied chunk does not show any connection to the declared workflow. This assessment is limited to the supplied code and does not establish what the rest of the skill does.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

所提供代码的实际功能是排版预览,与声明的检测、改写、复检及目标驱动重试用途存在实质差异。代码未显示任何检测、改写、复检、点数控制或版本保留逻辑;排版预览也不是该声明流程中可明确识别的辅助实现细节。

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The exposed prepare command accepts six feature types, and execute starts the selected feature through the remote API. Generation, formatting, and especially publishing are substantive capabilities beyond the declared detection–rewrite–recheck workflow. Publishing is not merely result reporting: the code explicitly supports starting publishing tasks and recognizes external draft/publish delivery modes, subject to approval checks. Credential management, task persistence, and local exports are supporting implementation details rather than independent mismatches. The supplied chunk can submit workflow specifications, but does not itself establish target-ratio retry logic or best-version selection; those behaviors may be delegated to the service.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The code supports the declared workflow through generic API calls, but also exposes substantial unrelated generation, formatting, and publishing functionality. Publishing is a distinct operational capability, not merely an implementation detail of AI detection and rewriting. The claimed retry and version-selection behavior might be implemented server-side, but it is not established by this code chunk. Credential storage, authenticated networking, polling, and result downloads are supporting implementation details and are not independently treated as mismatches.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied code performs document-format rendering. It contains no 优稿 integration, AI-content detection, rewriting, result verification, target-based retry logic, points-budget handling, or version preservation. Its vendor location suggests it could be a supporting dependency, but the supplied chunk does not establish a connection to the declared workflow. No network requests, file access, or code execution are shown; generated links and directives are output text.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
97% confidence
Finding

The supplied chunk has a materially different immediate purpose: Markdown-to-HTML rendering. It does not implement the declared workflow or demonstrate a connection to it. Its vendor-library location means it could be a supporting dependency in a larger skill, but that relationship is not established by the supplied evidence. This assessment is limited to the provided chunk; no network requests, local-file access, or other hidden side effects are shown.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The chunk contains no 优稿 integration, detection or rewriting calls, score-based retry logic, point-budget handling, or original/best-version preservation. Its implemented purpose is Markdown rendering. Although the vendor path suggests it could be a supporting dependency in a larger application, the supplied code establishes no connection to the declared workflow. This finding is limited to the provided chunk; it does not establish that the complete skill lacks the declared functionality.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The supplied code has a materially different immediate purpose: Markdown math support. It contains no 优稿 integration, detection or rewriting calls, target-score evaluation, retry controls, point-budget handling, or version retention. Although its vendor path suggests it could support a larger application, the supplied chunk does not establish a connection to the declared workflow. No network access, file access, or other suspicious side effects are shown.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

所提供代码仅注册 Markdown 渲染钩子、转换任务列表令牌并生成复选框 HTML,未显示调用优稿、检测文本、执行改写、按目标重试、控制点数或保存版本的逻辑。它位于 vendor 目录,可能是完整项目的辅助依赖,但本代码块没有体现与声明工作流的关联;就所提供代码本身而言,其主要用途与声明不符。未发现网络访问、文件访问或其他敏感操作。

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The supplied chunk implements local-file inclusion for a Markdown parser, a materially different function from the declared text-detection and rewriting workflow. No 优稿 integration, detection, rewriting, target-based retries, point budgeting, or version preservation appears here. Although this vendored component could support a larger application, the supplied evidence does not establish that connection. The directory and cycle checks constrain file access but do not align the demonstrated capability with the description.

Content

No source excerpt is available for this finding.

Direct Prompt Extraction

High
Category
System Prompt Leakage
Confidence
85% confidence
Finding

Skill contains instructions that could directly expose system prompts, internal rules, or hidden instructions to users or external parties.

Content

Scanner excerpt · scripts/vendor/mistune/list_parser.py (reported line 258)May include surrounding context.

python
]
    if "fenced_directive" in block.specification:
        rules.insert(1, "fenced_directive")
    return rules


def _match_list_item_break(

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: env, file_read, file_write, network.

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.

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/api.md (reported line 97)May include surrounding context.

md
## 各功能参数

| 功能      | input                                                                       | params                                                                                                                 |
| --------- | --------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------- |
| 文章生成  | brief 主题;可选 title、referenceText 参考正文、sourceImportId 已有导入记录 | style(最多200字)、prompt(6000字)、audience(200字)、length(100字)、titleRule(300字)                           |
| 文章改写  | text 原稿,可选 title;已有稿件用 manuscriptId 与 revision                  | 默认自然降 AI;style 可选深度降 AI、保留文风,prompt 补充要求                                                          |

The bundled helper exposes unrelated generation, formatting, workflow, and publishing execution capabilities beyond detection and iterative rewriting.

Medium
Category
Excessive Agency
Confidence
88% confidence
Finding

L021 and L317–L318 expose all six service features through the command line, and L134–L153 can start paid operations or external publishing rather than merely supporting the refine process. Although execution requires approval where applicable, these independently callable capabilities exceed the manifest's detection, rewriting, and rechecking purpose; the shared-client documentation explains their presence but does not make them necessary for this skill.

Content

No source excerpt is available for this finding.

Markdown include directives can incorporate additional local files beyond the manuscript being refined.

Medium
Category
Excessive Agency
Confidence
86% confidence
Finding

The include plugin resolves a document-specified filename, reads its contents at L070–L071, and incorporates them into parsed or rendered output. Although access is confined to the source directory and requires this optional plugin to be registered, content-directed inclusion of additional local files is not justified by the manifest's detection, rewriting, and rechecking workflow.

Content

No source excerpt is available for this finding.

The client exposes publishing operations unrelated to the declared detection-and-rewriting workflow.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The manifest limits this skill to detecting AI-generated text, rewriting it, and rechecking against a target ratio. L225–L234 implement previewing, starting, and retrieving publishing runs, while L300–L314 expose publishing account/manuscript queries and delivery decisions through the CLI. These are operational publishing capabilities, not implementation details needed for text refinement.

Content

No source excerpt is available for this finding.

The delivery instructions require a Chinese-language summary without offering a language choice.

Low
Category
Output Handling
Confidence
97% confidence
Finding

L46 specifies that report.txt is a Chinese summary, imposing a particular natural language on a required deliverable. The file provides neither user opt-in nor a documented locale-specific justification for this requirement.

Content

No source excerpt is available for this finding.

Scope Creep

Low
Category
Excessive Agency
Confidence
70% confidence
Finding

Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.

Content

Scanner excerpt · scripts/vendor/mistune/LICENSE (reported line 14)May include surrounding context.

text
* Neither the name of the creator nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.


THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

Scope Creep

Low
Category
Excessive Agency
Confidence
70% confidence
Finding

Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.

Content

Scanner excerpt · scripts/vendor/mistune/LICENSE (reported line 14)May include surrounding context.

text
* Neither the name of the creator nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.


THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · scripts/vendor/mistune/block_parser.py (reported line 108)May include surrounding context.

python
self.list_rules = list_rules
        self.max_nested_level = max_nested_level
        # register default parse methods
        self._methods = {name: getattr(self, "parse_" + name) for name in self.SPECIFICATION}

    def parse_blank_line(self, m: Match[str], state: BlockState) -> int:
        """Parse token for blank lines."""

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · scripts/vendor/mistune/inline_parser.py (reported line 118)May include surrounding context.

python
else:
            self.rules.append("softbreak")

        self._methods = {name: getattr(self, "parse_" + name) for name in self.rules}

    @property
    def max_nested_level(self) -> int:

Dynamic attribute access via getattr()

Low
Category
Dangerous Code Execution
Confidence
50% confidence
Finding

Dynamic getattr() with a non-literal attribute name can access arbitrary object attributes, potentially bypassing access controls.

Content

Scanner excerpt · scripts/vendor/mistune/plugins/__init__.py (reported line 48)May include surrounding context.

python
module_path, func_name = name.rsplit(".", 1)

    module = import_module(module_path)
    plugin = cast(Plugin, getattr(module, func_name))
    _cached_modules[name] = plugin
    return plugin