The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The README.md and installation instructions promote one-click installation via curl -fsSL https://raw.githubusercontent.com/OpenDemon/anyone-to-skill/master/install.sh | bash and a similar PowerShell command. These scripts download and execute code from a remote repository at runtime.
[COMMAND_EXECUTION]: Several Python scripts, including scripts/analyze_video.py, scripts/crawl.py, scripts/ingest.py, and scripts/distill.py, use subprocess.run to execute shell commands for tools like yt-dlp, ffmpeg, and pdftotext. These commands process external inputs such as YouTube URLs and file paths provided by the user.
[EXTERNAL_DOWNLOADS]: The skill fetches persona-specific SKILL.md files from various GitHub repositories under the OpenDemon organization (e.g., OpenDemon/elon-musk-skill) during the chat interface initialization.
[PROMPT_INJECTION]: The distillation pipeline (Mode 2) is vulnerable to indirect prompt injection. It ingests untrusted data from external sources like YouTube transcripts and user-uploaded chat logs to generate behavioral instructions in a new SKILL.md. Malicious content in these transcripts could influence the resulting agent's persona or instructions.
Ingestion points: YouTube transcripts via crawl.py, local files/chat logs via ingest.py.
Boundary markers: Uses simple delimiters like ---TRANSCRIPT--- in analyze_video.py.
Capability inventory: File system writes, shell command execution via subprocess.run.
Sanitization: Relies on LLM summarization and structured prompts but lacks robust sanitization of the raw transcript data.

anyone-to-skill