Skillv5.0.1

ClawScan security

YouTube Research Assistant · ClawHub's context-aware review of the artifact, metadata, and declared behavior.

Scanner verdict

BenignMar 7, 2026, 11:53 AM

Verdict: benign
Confidence: high
Model: gpt-5-mini
Summary: The skill's code, instructions, and required binaries are consistent with its stated purpose (fetching YouTube subtitles via yt-dlp and answering from stored transcripts); there are no unexplained credential or install demands, but note a few operational details to review before installing.
Guidance: Before installing, consider these points: (1) The skill runs yt-dlp as a subprocess which will make network requests to YouTube to fetch subtitles — ensure you are comfortable with that behavior and with yt-dlp being present on the host. (2) Transcripts and a small session/index file are stored locally under ~/.openclaw/workspace/skills/youtube-research-assistant/data for up to 24 hours; remove them if you need longer retention control. (3) The script calls yt-dlp with --no-check-certificates, which disables TLS certificate validation — review/modify that flag if you want stricter network security. (4) The SKILL.md mandates that all answers must come only from transcripts; the script enforces retrieving chunks but translation/language-detection is not implemented in the script itself — the agent will need to handle that. (5) As with any skill that executes subprocesses and writes files, only install if you trust the skill source; you can inspect/modify scripts locally before enabling. If you want tighter guarantees, request the author remove --no-check-certificates and add optional config to change the data directory or retention period.

Review Dimensions

Purpose & Capability: okName/description ask for transcript extraction and summarization; the skill only requires python3 and yt-dlp and provides a script that fetches subtitles, stores transcripts, and answers questions from them. The requested binaries and local storage are proportionate to the stated purpose.
Instruction Scope: noteSKILL.md strictly limits external calls to the single yt-dlp subtitle fetch and the included script follows that rule (it runs yt-dlp as a subprocess and otherwise works with local files). Two operational notes: (1) the script invokes yt-dlp with --no-check-certificates and --force-ipv4 which weakens TLS validation/network behavior — this is a security/operational choice to be aware of; (2) SKILL.md mentions language detection/translation but the script does not implement translation — the agent itself must perform language detection/translation as part of response generation.
Install Mechanism: okNo automated install step included; the skill is instruction + a local Python script. No remote archives or downloads are performed by the skill itself during install. The only runtime external tool is yt-dlp (expected).
Credentials: okThe skill requires no environment variables or credentials. It stores transcripts and a small session/index JSON under ~/.openclaw/workspace/skills/youtube-research-assistant/data which matches its function and does not request unrelated secrets or access.
Persistence & Privilege: okThe skill does not request always:true or elevated privileges. It persists transcripts and session state within its own skill folder and automatically cleans older transcripts (24h). This is appropriate for the stated functionality.