The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using sips, ffmpeg, and ffprobe for media processing tasks, such as cropping images to specific aspect ratios and converting video clips to GIFs.
[EXTERNAL_DOWNLOADS]: The skill references and depends on an external MCP server, note-mcp, hosted in a public GitHub repository (drillan/note-mcp).
[PROMPT_INJECTION]: The skill ingests untrusted user data, including article themes, target audiences, and media files, which are used to generate content and metadata for publication. This creates a surface for indirect prompt injection.
Ingestion points: User-provided themes, targets, goals, and video files are processed in SKILL.md (Step 1 and Step 6.5).
Boundary markers: The templates in templates/article.md use basic placeholders but do not include explicit security delimiters to isolate user-generated content from system instructions.
Capability inventory: The skill can perform network operations via note-mcp tools (creating drafts, uploading images, publishing articles) and execute shell commands via ffmpeg and sips.
Sanitization: No explicit sanitization or validation of user-provided strings or file metadata is mentioned before interpolation into prompts or shell commands.
[SAFE]: Security is enhanced by mandatory manual verification steps (Step 2 and Step 8), ensuring the user reviews the article structure and provides explicit permission before any content is published live.

note-article