ClawHub
Back to skill

Security audit

video-clip-skill

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly matches its video-clipping purpose, but its “local” workflow can upload extracted audio to Groq without an explicit confirmation step.

Review this skill before installing if you expect all processing to stay on your machine. It is most appropriate for non-sensitive YouTube clips; avoid private, confidential, or copyrighted audio unless you are comfortable with Groq transcription uploads, and consider changing the fallback to require explicit confirmation before any audio is sent externally.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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 (6)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding
The code chunk’s actual purpose is narrow: subtitle format conversion/styling from VTT to ASS with karaoke timing and optional display of pre-supplied translated subtitles. This partially overlaps with the declared subtitle-generation/karaoke aspect, but the declared primary purpose is a broader local YouTube clipping pipeline involving yt-dlp, ffmpeg, highlight extraction, and optional transcription/translation. None of those broader behaviors appear in this code. Because the implementation shown materially differs from the declared main function and omits most advertised capabilities, this is a description-behavior mismatch.

Agent Config Directory Access

High
Category
Agent Snooping
Content
The ASS karaoke generator is bundled with this plugin. Locate it once at the start (this only searches for the plugin's own bundled file):

```bash
ASS_SCRIPT=$(find ~/.claude/plugins -path '*/clip-local/*/scripts/ass-karaoke.py' 2>/dev/null | head -1)
```

## Auto-highlight mode
Confidence
88% confidence
Finding
Searching `~/.claude/plugins` gives the skill visibility into the agent's configuration/plugin directory, which expands access beyond the immediate working context and can expose local environment structure or other installed plugins. Although the stated intent is to locate a bundled script, broad home-directory traversal is riskier than necessary and can become dangerous if the path resolution is influenced by unexpected files or symlinks.

Lp3

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding
Without declared permissions the skill's intent is opaque and cannot be validated.

Vague Triggers

Medium
Confidence
92% confidence
Finding
The manifest says the skill triggers when the user wants 'local video clipping, highlight extraction, or subtitle generation,' which is a broad natural-language description rather than a specific invocation boundary. Phrases like 'highlight extraction' or 'subtitle generation' can arise in many contexts and the file provides no negative examples or narrowing constraints to prevent accidental activation.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The skill sends user-derived audio content to Groq for transcription, but the description does not prominently warn that media may leave the local machine. For a tool marketed as 'local' clipping, this creates a meaningful privacy and trust risk because users may reasonably assume processing stays local unless clearly told otherwise.

External Transmission

Medium
Category
Data Exfiltration
Content
If yt-dlp finds no auto-subs and user has `GROQ_API_KEY` set:

1. Download audio: `yt-dlp -f ba -x --audio-format mp3 --postprocessor-args 'ffmpeg:-ac 1 -ar 16000 -b:a 64k'`
2. Transcribe: `POST https://api.groq.com/openai/v1/audio/transcriptions` with `model=whisper-large-v3`, `response_format=verbose_json`
3. Convert segments to VTT

If `GROQ_API_KEY` is not set, inform the user that no subtitles are available and ask how to proceed (clip without subs, or set the key).
Confidence
95% confidence
Finding
This skill transmits audio extracted from the user's target video to an external transcription service. Even if functionally justified, audio may contain sensitive or copyrighted material, and the 'local' positioning makes undisclosed off-device transfer more risky in context.

Static analysis

No suspicious patterns detected.