The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from external Douyin links and user-provided video files, exposing it to potential injection attacks.
Ingestion points: External URLs and local file paths are ingested in SKILL.md and passed to the scripts/transcribe.js script.
Boundary markers: The instructions lack explicit boundary markers or sanitization directives to handle adversarial input within user-provided strings.
Capability inventory: The skill uses browser automation and shell command execution through the exec tool and internal script calls.
Sanitization: There is insufficient sanitization of input strings in scripts/transcribe.js before they are interpolated into shell command templates for ffmpeg, ffprobe, and yt-dlp.
[COMMAND_EXECUTION]: The skill executes shell commands using Node.js to call binaries like ffmpeg and yt-dlp. Because user-controlled strings such as URLs and filenames are incorporated into these commands, it creates a surface for shell command injection.
[EXTERNAL_DOWNLOADS]: The skill downloads media from Douyin and interfaces with well-known AI service providers, specifically Groq (api.groq.com) and OpenAI (api.openai.com), for transcription and text processing. These external network operations are consistent with the skill's intended functionality.

douyin-transcribe