Security audit

Showcase Video Builder

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a local ffmpeg helper, but its bundled shell script handles filenames unsafely enough that untrusted image names could change the ffmpeg command being run.

Review this before installing if you may process screenshots or images from other people. The skill should be limited to trusted local image directories, or the script should be fixed to use shell arrays and quoted arguments before use. It does not show evidence of hidden persistence or credential theft, but the current script can run unintended ffmpeg options through malicious filenames.

A.I.G

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: scripts/build_showcase.sh:10
Finding: FFmpeg Argument Injection Through Unquoted Input Construction
Content: View full analysis
Remediation: View remediation
&2 exit 1 fi concat="" for ((j = 0; j < i; j++)); do concat+="[v$j]" done "$FFMPEG" "${inputs[@]}" \ -filter_complex "${filter} ${concat}concat=n=$i:v=1:a=0[out]" \ -map "[out]" -c:v libx264 -pix_fmt yuv420p -r "$FPS" -- "$OUTPUT" ``` Additional hardening measures: 1. Resolve `FFMPEG` to an approved executable path or reject values containing whitespace instead of accepting an arbitrary command-like string. 2. Treat screenshot directories and filenames as untrusted input. 3. Run FFmpeg in a sandbox with minimal filesystem permissions and no network access when processing untrusted media. 4. Restrict FFmpeg protocols where supported, allowing only those required for local image processing. 5. Validate that each discovered input is a regular local file and an expected image type before invoking FFmpeg. 6. Fail explicitly when no PNG files are found rather than generating a malformed filter graph. 7. Add regression tests using filenames containing spaces, wildcard chara ...[truncated 102 chars]

SkillSpector

Vulnerability Patterns

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
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (1)

Tp4

High

Category: MCP Tool Poisoning
Confidence: 90% confidence
Finding: The code generally aligns with the broad idea of building a showcase video from static assets using ffmpeg, so its primary purpose is related. However, the declared description claims support for screenshots, avatars, and text overlays to create polished demo videos, while the supplied code only loops over PNG files and concatenates them into a basic slideshow with fades. There is no implementation for avatars, text overlays, richer composition, or handling multiple asset types. This is a meaningful description-to-behavior mismatch because the declared capabilities are materially broader than what the code actually does.
Content: No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.