Wjs Segmenting Video

• Outcome: Review
• Latest audit: May 29, 2026 · 9:39 PM UTC
• Version: 0.1.0

Overview

The core video-cutting workflow is understandable, but the package includes under-disclosed post-production helpers and an unsafe local ffmpeg lookup that users should review before installing.

Install only if you are comfortable reviewing and controlling which helper scripts run. Use the core segment.py and --no-burn SRT slicing path for the declared workflow, set FFMPEG to a trusted binary, avoid relying on /tmp/ff_bin/ffmpeg, and do not run make_cover.py or prepend_intro.py unless you explicitly want AI cover generation, external frame sharing, or intro-card compositing.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Behavioral AST: exec() Call, eval() Call, Dynamic Import
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration

Findings (6)

subprocess module call

Medium

• Category: Dangerous Code Execution

Content

f.write(f"file '{body_path.resolve()}'\n")
        list_path = Path(f.name)
    try:
        proc = subprocess.run(
            [ffmpeg, "-y", "-f", "concat", "-safe", "0",
             "-i", str(list_path),
             "-c", "copy", "-movflags", "+faststart", str(out_path)],

• Confidence: 85% confidence
• Finding: proc = subprocess.run( [ffmpeg, "-y", "-f", "concat", "-safe", "0", "-i", str(list_path), "-c", "copy", "-movflags", "+faststart", str(out_path)],

Description-Behavior Mismatch

High

• Confidence: 95% confidence
• Finding: The file implements AI cover generation and title-card behavior even though the skill manifest says this skill only cuts/crops raw clips and emits per-clip SRTs. This scope mismatch is dangerous because it can cause the agent to perform undeclared external actions, transmit user media to another tool/provider, and produce outputs the user did not consent to within this skill's stated contract.

Context-Inappropriate Capability

Medium

• Confidence: 92% confidence
• Finding: The script invokes an external AI image editing tool and sends prompts plus reference frames to it, which exceeds a segmentation-only workflow and introduces a new data-flow boundary. In this context, that is security-relevant because user video frames may be disclosed to an external provider without being clearly justified by the declared skill purpose.

Intent-Code Divergence

Medium

• Confidence: 90% confidence
• Finding: The module documentation states covers are used as prepended intro title-cards, directly contradicting the manifest's claim that this skill only produces raw clips for downstream post-production. This hidden behavioral expansion is risky because it changes media outputs and workflow semantics in ways users and orchestrators may not expect or authorize.

Description-Behavior Mismatch

High

• Confidence: 89% confidence
• Finding: The skill metadata says the skill should only cut and crop raw clips for handoff, but this script adds title-card intros and changes the media package beyond declared scope. In an agent setting, scope expansion is dangerous because users and orchestrators may trust the skill to perform only minimally transformative operations, while this code silently alters output semantics and downstream behavior.

Context-Inappropriate Capability

Medium

• Confidence: 82% confidence
• Finding: The script preferentially operates on *_burned.mp4 subtitle-burned variants, which conflicts with the stated purpose of outputting raw clips and per-clip SRTs for later post-production. In a multi-skill workflow, touching burned derivatives can break separation-of-duties assumptions, produce irreversible outputs, and bypass intended downstream review/editing steps.

VirusTotal

59/59 vendors flagged this skill as clean.

Static analysis

No suspicious patterns detected.