The Agent Skills Directory

[COMMAND_EXECUTION]: Multiple scripts, including scripts/main.py, scripts/convert.py, scripts/download_douyin.py, and scripts/core/downloader.py, use subprocess.run to execute shell commands such as ffmpeg, yt-dlp, and pandoc.
[PRIVILEGE_ESCALATION]: The environment setup script scripts/check_environment.py attempts to acquire root privileges using sudo to install the ffmpeg package via apt-get or dnf on Linux systems.
[CREDENTIALS_UNSAFE]: The script scripts/fetch_bilibili_subtitle.py attempts to extract Bilibili session cookies (SESSDATA) directly from local browser databases (Chrome, Firefox, Edge, Safari) using yt-dlp or the browser_cookie3 library. This involves accessing sensitive user data without explicit user interaction in the script logic.
[EXTERNAL_DOWNLOADS]: The skill performs runtime installation of Python packages (yt-dlp, fpdf2) and system binaries (ffmpeg, pandoc) if they are not detected on the host system. This happens automatically during the execution of scripts/check_environment.py and scripts/convert.py.
[DATA_EXFILTRATION]: scripts/transcribe_api.py extracts audio from user-provided videos and uploads it to an external API endpoint (base_url defined in ~/.easyclaw/easyclaw.json) for transcription.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from video transcripts and external web pages via browser automation. The resulting content is structured and analyzed by the AI agent, creating a surface for indirect prompt injection if the source video metadata or transcripts contain malicious instructions.
Ingestion points: scripts/main.py (via scripts/fetch_bilibili_subtitle.py and scripts/transcribe_api.py), SKILL.md (via browser tool access).
Boundary markers: Absent. Transcripts are processed as raw text for analysis.
Capability inventory: Subprocess execution for video processing, PDF conversion, and system package installation.
Sanitization: None. The skill relies on ASR output and raw web content.

short-video-replicator