The Agent Skills Directory

[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface due to its reliance on untrusted web data for content generation.
Ingestion points: Untrusted data enters the context via WebSearch and WebFetch tools during Phase 2 of the execution flow in SKILL.md.
Boundary markers: The prompt template in assets/note-template.md and the instructions in SKILL.md lack explicit delimiters or safety instructions to isolate fetched content from the agent's logic.
Capability inventory: The skill possesses capabilities for filesystem writes (saving reports in Phase 5) and shell command execution (running media scripts in Phase 3).
Sanitization: No sanitization, validation, or escaping of retrieved web content is performed before interpolation into the final document or command-line arguments.
[EXTERNAL_DOWNLOADS]: The documentation instructs the agent and user to install additional helper skills (zimage-skill and media-downloader) from the vendor's GitHub repository. These are documented as vendor-owned resources.
[COMMAND_EXECUTION]: The skill orchestrates the execution of shell commands using system utilities like ffmpeg and yt-dlp, as well as Python-based CLI tools from the developer's auxiliary skills, incorporating parameters such as URLs and keywords derived from external research.

chuinb