Back to skill

Security audit

xiaohongshu-create

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a straightforward Xiaohongshu publishing workflow, but it can post publicly and also records results to Feishu without clearly scoped user confirmation or disclosure.

Install only if you are comfortable with an agent operating a logged-in Xiaohongshu creator session. Before use, require the agent to show the exact post, destination account, and final action, and do not allow the Feishu recording step unless you have explicitly approved the workspace, library, and fields to be recorded.

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
Findings (1)

other

Warning
Location
SKILL.md:58
Finding
Undisclosed External Data Handling Through Feishu## Vulnerability Details **File Location**: `SKILL.md`, lines 58–61 **Vulnerability Type**: Unnecessary external data handling **Risk Level**: Medium ### Evidence ```markdown ### 步骤 5:确认成功 1. 等待页面提示"发布成功" 2. 记录到飞书"已发布内容库" ``` ### Technical Analysis After publishing content to Xiaohongshu, the skill instructs the agent to record publication information in a Feishu repository named “Published Content Library.” This secondary external data transfer is not necessary to fulfill the primary publishing request. The instruction does not specify the authorized Feishu workspace, permitted destination, fields to be copied, applicable data-minimization rules, or whether the user has consented to the transfer. The additional side effect is also absent from the workflow documented in `README.md`, making it less visible to users reviewing the skill’s stated behavior. This issue does not match classifications T01–T09 because the available evidence does not demonstrate privilege escalation, persistent memory poisoning, malicious code, dependency compromise, or tool hijacking. It is classified as an external data-handling risk. ### Attack Path 1. A user asks the agent to publish approved content to Xiaohongshu. 2. The agent invokes the skill and publishes the content through the Xiaohongshu Creator Platform. 3. After receiving a successful-publication confirmation, the agent follows `SKILL.md` line 61. 4. The agent writes publication information to an unspecified Feishu “Published Content Library.” 5. User content or publication metadata is thereby duplicated into a third-party workspace without an explicit, operation-specific authorization requirement. ### Impact Assessment The instruction could cause unauthorized disclosure or duplication of published content and associated metadata into an unspecified Feishu workspace. It may also permit unintended modification of an organizational content repository using whatever Feishu access is already available to the agent. T ...[truncated 283 chars]
Remediation
## Remediation Suggestions 1. Remove the Feishu-recording step if it is not essential to the user-requested Xiaohongshu publication. 2. If archival is required, request explicit user consent before every external transfer. 3. Clearly identify the authorized Feishu tenant, workspace, application, and destination repository. 4. Define the exact fields that may be recorded and apply data minimization; avoid copying full content when a publication URL and status are sufficient. 5. Validate that the destination belongs to the intended user or organization before writing. 6. Require a preview and confirmation of the destination and data before creating or updating a record. 7. Document this side effect in `README.md` so the public workflow accurately describes all external services and data writes. 8. Log the transfer outcome without exposing sensitive content, and provide a clear way to disable archival.
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 Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (7)

Missing User Warnings

Medium
Confidence
98% confidence
Finding
The README describes a workflow that culminates in publishing to Xiaohongshu but does not warn users that the skill may send content to an external service. This reduces informed consent and increases the risk of privacy, reputational, or operational harm if users provide sensitive draft content or do not realize the action may result in a public post.

Vague Triggers

Medium
Confidence
96% confidence
Finding
The trigger phrase “发布小红书” is broad enough to overlap with ordinary conversation, which can cause the skill to activate unexpectedly. Because this skill performs a publishing workflow to an external platform, accidental invocation could lead to unintended posting actions or movement toward publication without sufficiently explicit user intent.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The skill describes externally impactful actions—posting content to Xiaohongshu and recording results to Feishu—without an explicit confirmation gate before execution. In an agent setting, this creates a real risk of accidental publication or unintended cross-service data sharing, which can immediately affect the user’s public presence and leak content or metadata to another platform.

Description-Behavior Mismatch

Medium
Confidence
94% confidence
Finding
The skill’s stated purpose is publishing to Xiaohongshu, but it also instructs writing publication data into Feishu, introducing a second external side effect outside the core scope. This can cause unanticipated data disclosure or action chaining, especially if published content, links, or metadata are recorded without the user explicitly agreeing to cross-platform transfer.

Context-Inappropriate Capability

Medium
Confidence
91% confidence
Finding
Recording to a Feishu content library is not justified by the described Xiaohongshu publishing function, so it represents unnecessary capability expansion. Unnecessary external integrations increase the attack surface and the chance that user content, account identifiers, or publishing history are copied into another system without clear consent.

Natural-Language Policy Violations

Low
Confidence
71% confidence
Finding
The entire skill description is written only in Chinese and presents a fixed operating flow without indicating any language choice or documenting that the skill is intentionally limited to Chinese-speaking users. This can violate language/locale policy when a specific language is effectively enforced without user opt-in.

Natural-Language Policy Violations

Low
Confidence
81% confidence
Finding
Natural-language policy violations include forcing a specific language or locale without user opt-in. Here, the skill name/description and invocation phrases are exclusively in Chinese, and the file does not indicate that the skill is region-specific or offer alternative language use.

Static analysis

No suspicious patterns detected.