Back to skill

Security audit

品牌热点营销助手

Security checks for vulnerabilities and agentic risk

Overview

The skill is a coherent paid Xiaohongshu marketing assistant, but it needs Review because it handles App-Keys insecurely and includes a real-looking key in its instructions.

Install only if you are comfortable giving this provider an App-Key and brand names for processing. Treat the published example key as unsafe, avoid pasting long-lived credentials into normal chat when possible, confirm the real HTTPS endpoint before use, and manually post to Xiaohongshu as the skill itself recommends.

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 (1)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:107
Finding

Hard-Coded App-Key in a Plaintext Local API Request

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 107-118
Vulnerability Type: Hard-coded credential and insecure credential transmission
Risk Level: Medium

Vulnerable Code

markdown
**Base URL**: `http://127.0.0.1:8000` (replace with the actual domain in production)

**POST** `/gateway/vane_marketing/content`

### Request Example

```bash
curl --location --request POST 'http://127.0.0.1:8000/gateway/vane_marketing/content' \
--header 'App-Key: S6YVomNq2PYl4bnjGYXDeaM1J9jhojwv' \
--header 'Content-Type: application/json' \

Technical Analysis

The request example embeds a credential-shaped App-Key directly in reusable documentation. If the value is valid, copying the example can expose it through shell history, process inspection, terminal recordings, logs, documentation mirrors, or source-control history.

The documented content-generation request also sends the App-Key to an HTTP service on 127.0.0.1:8000. Although loopback traffic is not ordinarily exposed to remote network interception, HTTP provides no server authentication. Any local process able to bind to or control port 8000 could impersonate the expected gateway and capture the key. The documentation does not prescribe endpoint authentication, key scoping, secret redaction, secure storage, or validation of the process receiving the credential.

The same skill instructs users to submit their App-Key during the conversation and then forwards it to external or local API endpoints. This can cause user credentials to enter conversation histories and request logs. The audit cannot establish whether the embedded example value remains active, but placing a realistic credential value in published instructions is independently an insecure practice.

Attack Path

  1. A user or agent copies the documented command without replacing the embedded App-Key.
  2. The key is retained in shell history, execution logs, chat history, or other moni ...[truncated 914 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove the literal App-Key from all examples and replace it with an unmistakable placeholder such as ${VANE_APP_KEY} or REPLACE_WITH_YOUR_APP_KEY.
  2. Treat the published value as potentially compromised. Verify whether it is real and, if so, revoke and rotate it immediately.
  3. Obtain credentials from a protected secret store or runtime environment injection rather than asking users to disclose complete keys in ordinary conversation.
  4. Redact keys from chat transcripts, command output, application logs, error messages, telemetry, and audit records.
  5. Replace the unspecified HTTP gateway with a documented HTTPS endpoint whose certificate and hostname are validated.
  6. If a loopback gateway is required, authenticate the local service, verify its identity, restrict port access, and avoid forwarding the primary billing credential to it.
  7. Prefer scoped, short-lived tokens dedicated to the content-generation operation instead of long-lived account-level App-Keys.
  8. Add key rotation, revocation, rate limits, anomaly detection, and quota alerts to reduce the consequences of credential exposure.
  9. Update examples to load the secret safely without exposing it in command-line arguments or shell history.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (3)

Tp2

High
Category
MCP Tool Poisoning
Confidence
85% confidence
Finding

Mixing characters from multiple Unicode scripts in a single identifier is a common technique to create visually ambiguous tool names.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest description specifies generating Xiaohongshu content in Chinese and the skill instructions/examples consistently assume Chinese output and interaction, but the file does not offer the user a language choice or explicit opt-in. This is a natural-language locale policy concern because it imposes a specific language by default rather than documenting it as an optional or justified constraint.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
99% confidence
Finding

The skill instructs the agent to collect a user App-Key and transmit it in an HTTP header, and the example also hardcodes what appears to be a real secret in documentation. Even though the sample target is localhost, this normalizes secret handling by the agent, risks credential exposure in logs/examples, and could leak or misuse user credentials if deployed against a real endpoint or if the local service is controlled by another process.

Content

Scanner excerpt · SKILL.md (reported line 114)May include surrounding context.

请求示例

bash
curl --location --request POST 'http://127.0.0.1:8000/gateway/vane_marketing/content' \
--header 'App-Key: S6YVomNq2PYl4bnjGYXDeaM1J9jhojwv' \
--header 'Content-Type: application/json' \
--data-raw '{