Back to skill

Security audit

Weiyun Toolkit Free

Security checks for vulnerabilities and agentic risk

Overview

This skill mostly matches cloud file management, but it handles Weiyun tokens/cookies while leaving endpoint and download URL validation unclear.

Review before installing. Only use this skill for explicit Weiyun file tasks, provide tokens through environment variables, confirm every local file path before upload, avoid custom MCP endpoints unless you fully trust them, and validate download URLs/redirects against trusted Tencent/Weiyun hosts before sending cookies.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (2)

T09 · Insecure Skill Coding Practices

Warning
Location
SKILL.md:132
Finding

Authentication Cookie May Be Exposed Through an Unvalidated Download URL

Content
View full analysis
-b "" "" ``` ``` ### Technical Analysis The Skill directs the Agent to pass an authentication cookie to a download URL supplied by the Weiyun MCP response. It also enables automatic redirect following through `curl -L`. The instructions do not require validation of the initial URL or each redirect destination against an explicit trusted-host allowlist. Merely requiring an HTTPS URL would not be sufficient because an attacker-controlled host can also provide valid HTTPS. If the MCP response is compromised, spoofed, or manipulated, it could provide an attacker-controlled URL or redirect chain. Depending on `curl` cookie processing and the supplied cookie format, sensitive cookie material may be transmitted outside the intended Weiyun service. At minimum, the command allows untrusted service output to control a credential-bearing network request. ### Attack Path 1. An attacker compromises or impersonates the MCP endpoint, or otherwise manipulates the response returned to the Agent. 2. The malicious response supplies an attacker-controlled `https_download_url`, either directly or through an HTTP redirect. 3. The Agent follows the Skill instructions and executes `curl` with the returned cookie and URL. 4. The request follows redirects because `-L` is enabled. 5. Cookie material may be disclosed to attacker-controlled infrastructure, enabling unauthorized use within the scope granted by that cookie. ### Impact Assessment Successful exploitation may disclose a temporary or persistent download authentication cookie. The resulting privileges depend on the cookie's server-side scope and lifetime. Potential impact includes unauthorized acc ...[truncated 257 chars]
Remediation
View remediation

T08 · Insecure Dependencies

Note
Location
SKILL.md:231
Finding

Unpinned Python Dependency Installation Creates Supply-Chain Risk

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (6)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The activation guidance is overly broad and includes unrelated data analysis, reporting, visualization, team, and workflow scenarios, which can cause an agent to select this skill in contexts where file access, upload, download, or exec capabilities are unnecessary. In a skill that declares read/write/exec tools and handles tokens plus local files, over-selection increases the attack surface and the chance of unintended credential or file operations.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill advertises a narrowly scoped free edition for browse/download/upload, but the execution sections describe generic create, modify, reset, import, export, save, and conversion operations without tight linkage to the stated tool set. In an agent setting, this scope ambiguity can cause the skill to be invoked for broader file-changing tasks than intended, increasing the chance of unauthorized local or remote actions via read/write/exec-enabled workflows.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The capability coverage line enumerates many fragmented, everyday keywords and broad automation phrases without clear trigger boundaries. In an agent ecosystem, such keyword stuffing can inappropriately attract prompts outside the intended scope, leading the agent to expose cloud-file and local-exec functionality in unrelated tasks.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The documentation exposes a user-controllable --mcp_url parameter while later claiming only whitelisted weiyun/qq endpoints are accepted. If the script actually honors arbitrary URLs, an attacker could direct the agent to send the MCP token, file metadata, or uploaded content to an attacker-controlled endpoint, defeating the stated privacy and trust assumptions.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill is described elsewhere as a 微云 file-management toolkit for listing, downloading, and uploading files. The 'Use when' guidance instead recommends it for data analysis, report generation, statistical insights, and data visualization, which are unrelated higher-level analytical capabilities not supported by the described operations.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
75% confidence
Finding

The FAQ states that on Windows the user 'must' switch the console to UTF-8 and presents the guidance only in Chinese-language terms, without offering locale flexibility or clarifying that this is a technical workaround rather than a language requirement. This can be interpreted as imposing a specific locale handling expectation without user opt-in.

Content

No source excerpt is available for this finding.