Security audit

ZeroCut AI Video

Security checks for vulnerabilities and agentic risk

Overview

This skill is a straightforward ZeroCut command reference for media/document generation and sandboxed conversions, with disclosed local output behavior.

Install this if you want agents to use ZeroCut for media generation and sandboxed ffmpeg or pandoc work. Before running commands, specify trusted input resources and explicit output filenames or directories, since generated or converted files may be written into the current workspace and missing output directories may be created.

A.I.G

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

SkillSpector

Vulnerability Patterns

Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (5)

Vague Triggers

Medium

Confidence: 93% confidence
Finding: The top-level description says to invoke the skill whenever the user needs to generate media, run ffmpeg/pandoc, sync resources, or save outputs, which is broad enough to match many ordinary requests without requiring explicit user intent for tool execution. In an agent setting, overbroad activation increases the chance of unintended command execution, file syncing, or local writes, especially because this skill can fetch remote resources and write files automatically.

Vague Triggers

Medium

Confidence: 93% confidence
Finding: The 'When To Invoke' section lists broad categories like generating media, running ffmpeg/pandoc, syncing resources, and saving outputs, but does not define boundaries or safety preconditions. This can cause the agent to activate the skill for loosely related requests and perform operational actions when the user may only want guidance, creating risk of unintended processing, network fetches, or file creation.

Missing User Warnings

Medium

Confidence: 89% confidence
Finding: The ffmpeg behavior states that output files are auto-downloaded from the sandbox to the local current directory, but the skill does not warn that execution will cause a local filesystem write. In agent workflows, silent local writes can surprise users, overwrite expected workspace contents, or persist unreviewed data to disk without explicit approval.

Missing User Warnings

Medium

Confidence: 89% confidence
Finding: The pandoc behavior similarly says output is auto-downloaded to the local current directory without an explicit warning about local file creation. Because pandoc conversions often involve user-supplied inputs and generated artifacts, this can lead to unintended persistence of files or overwriting of expected local outputs if invoked automatically.

Missing User Warnings

Medium

Confidence: 91% confidence
Finding: The skill says output paths are resolved absolutely from the current working directory and that missing parent directories are created automatically, but it provides no safety warning or restriction on where files may be written. Automatic directory creation amplifies the impact of an unintended or attacker-influenced path by turning a simple output request into persistent filesystem modification beyond an already-existing safe workspace.

Static analysis

No suspicious patterns detected.