Back to skill

Security audit

multimodal-gen

Security checks for vulnerabilities and agentic risk

Overview

This skill is a Chinese-language multimodal prompt helper with an optional local learning log; it does not show hidden execution, exfiltration, or destructive behavior.

Install only if you are comfortable with a Chinese-language workflow and with the optional learner storing local preference and error notes in learned_patterns.json. Do not record sensitive prompts, personal data, secrets, or confidential business context in the learner notes unless you manage that file's retention and access yourself.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (10)

Tp4

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The documented purpose centers on multimodal generation and transcription, but the skill also includes a self-learning/persistence subsystem that records user preferences, failures, and reflections to disk without clearly declaring that behavior in the primary description. This mismatch is dangerous because it defeats user and reviewer expectations, hiding data collection and persistence behind an unrelated capability surface.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The file is a generic cross-skill learning and telemetry component, which does not match the declared multimodal generation purpose of the skill. In this context, capability drift is dangerous because it introduces unrelated persistent behavior tracking into a media-generation skill, increasing the chance of undisclosed data collection and making review/auditing harder.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
80% confidence
Finding

The skill advertises commands that write learning data to disk (learned_patterns.json) but does not declare any tool scope or permissions boundary. In an agent environment, undocumented file-write capability weakens least-privilege controls and can allow unexpected persistence of user-derived data or future abuse through expanded write behavior.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The self-learning commands explicitly persist user preference notes and failure details to disk, yet the skill provides no user-facing warning, consent flow, retention policy, or minimization guidance. That creates a privacy and compliance risk because sensitive personal preferences, usage traces, or error context may be stored locally without the user's informed approval.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The natural-language interface, usage instructions, and output strings are written exclusively in Chinese, which imposes a locale choice on users. The file does not indicate that this is optional, configurable, or required for a region-specific purpose.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

User preference and usage-related data are written to disk without any user-facing warning, consent flow, or indication of retention. This is dangerous because users may unknowingly provide prompts, notes, or preferences that become persistent local artifacts, creating privacy and compliance risk even if no external exfiltration occurs.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The code persistently stores usage history, error notes, recent operations, and user preferences in a shared skill directory model, despite the skill being described as multimodal content generation. This creates an unjustified telemetry surface that can capture sensitive prompts, notes, and preferences over time, enabling privacy leakage and cross-context profiling if the data is accessed by other components or users.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The module docstring, CLI help text, and usage examples are all presented only in Chinese, which imposes a specific language on users. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale restriction is clearly documented and justified, which it is not here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The command-line help and printed success message are user-facing strings written only in Chinese, with no mechanism for selecting another language. This constitutes a locale/language policy issue because the skill forces one language by default without user choice or documented justification.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
72% confidence
Finding

The text directs the user to indicate a language or dialect for speech transcription, but the document does not state that the skill supports user choice of language/locale or explain any locale constraints. This can be a language-policy concern if the skill implicitly expects a particular locale behavior without explicit opt-in or documented justification.

Content

No source excerpt is available for this finding.