Skillv1.0.0

ClawScan security

Youtube Video Analyzer · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

SuspiciousFeb 15, 2026, 1:33 AM

Verdict: suspicious
Confidence: medium
Model: gpt-5-mini
Summary: The skill's declared dependencies and extraction steps match its YouTube-analysis purpose, but the multimodal 'image analysis' phase is underspecified and the install kind is nonstandard — leaving room for unexpected behaviour (e.g., uploading frames to external services).
Guidance: This skill appears to legitimately implement video download, transcript extraction, and frame capture — those parts are coherent. The main risk is the unspecified image-analysis step: before installing, confirm where and how the extracted frames will be processed. Ask the publisher or inspect the full SKILL.md for any commands or API calls that would upload images or send data to external endpoints. Also verify what 'uv' install means in your environment and ensure yt-dlp will be fetched from a trusted package source. If you plan to analyze private or sensitive videos, run the skill in a restricted environment (sandbox), limit how many frames are saved, and ensure that your agent or any invoked tooling is not configured to forward files to third parties. If you need higher assurance, request the skill author provide the missing analysis code or a clear statement of which image-analysis services (if any) will be used.

Review Dimensions

Purpose & Capability: okThe name/description align with the requested binaries and the declared install of yt-dlp. ffmpeg, python3, curl and yt-dlp are exactly what you'd expect for downloading videos, extracting transcripts, and producing frames.
Instruction Scope: concernThe SKILL.md provides concrete, low-level steps for metadata retrieval, subtitle extraction, video download, and frame extraction — all scoped to the stated purpose. However the critical 'Multimodal analysis' step is vague: it says 'Read each frame image, combine with transcript, Generate structured output' but does not specify how or where images are analyzed. That vagueness gives an agent broad discretion, which could lead to unexpected actions such as calling external image-analysis APIs or uploading frames to third-party endpoints.
Install Mechanism: noteThe install spec installs yt-dlp via an installer kind labeled 'uv'. yt-dlp itself is an expected package for this use case, but the installer kind 'uv' is nonstandard/ambiguous in the provided metadata. If 'uv' maps to a known, audited package source in your environment this is low risk; if it downloads code from an untrusted host, it would be higher risk. No direct remote-download-from-arbitrary-URL pattern was found.
Credentials: okNo environment variables, credentials, or config paths are requested. The skill does not ask for unrelated secrets or system config access, which is proportionate to its stated purpose.
Persistence & Privilege: okalways:false and default invocation settings are used. The skill does not request persistent system-wide privileges or modifications to other skills. It writes temporary files only to a per-run temp directory.