The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The script scripts/download_tiktok.py performs dynamic code loading by modifying sys.path to include a directory provided via the TIKTOKDOWNLOADER_ROOT environment variable or the --tiktokdownloader-root argument. It then imports a module (src.link.extractor) from that path. This allows execution of arbitrary Python code from a computed or user-provided directory.
[COMMAND_EXECUTION]: The skill executes the external utility yt-dlp using subprocess.run to handle video downloads and metadata extraction.
[EXTERNAL_DOWNLOADS]: The skill fetches data from tikwm.com (an unofficial TikTok API) and downloads media content from various TikTok-related domains using urllib.request.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted metadata (titles, descriptions, and tags) from TikTok videos and stores them in download_manifest.json without sanitization.
Ingestion points: scripts/download_tiktok.py (via request_json from tikwm.com and read_json from yt-dlp sidecar files).
Boundary markers: None. Untrusted strings are placed directly into the JSON manifest.
Capability inventory: Subprocess execution (yt-dlp), network access (urllib), and file system writes.
Sanitization: Filenames are sanitized via regex, but the textual content of titles and descriptions is not filtered for embedded instructions.

bggg-tiktok-downloader