The Agent Skills Directory

[COMMAND_EXECUTION]: The skill generates and executes shell commands using user-provided paths or URLs stored in a variable.
Evidence: In SKILL.md, Step 2 provides specific bash snippets for technical inspection: ffprobe -v error -show_entries format=duration,size:stream=width,height,r_frame_rate -of json "$VIDEO" and ffmpeg -y -i "$VIDEO" -vf fps=1 /tmp/video-frames/frame-%03d.jpg. The reliance on the $VIDEO variable without instructions to sanitize the input allows for potential command injection if a maliciously crafted path is provided.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from various external sources and possesses capabilities that could be exploited if that data contains malicious instructions.
Ingestion points: SKILL.md (Step 1) specifies that the skill accepts local paths, uploaded files, URLs, and can inspect source HTML/CSS/JS or local pages connected to the video.
Boundary markers: The final output is structured within a fenced text block as defined in Step 5 and the references/superprompt-template.md file, but there are no boundary markers or "ignore instructions" warnings for the data ingestion and analysis phase.
Capability inventory: The skill utilizes subprocess execution for ffmpeg and ffprobe, writes files to the local directory /tmp/video-frames, and may perform git commit operations as mentioned in Step 6.
Sanitization: There is no evidence of input validation, escaping, or sanitization for the external data being processed.

video-to-superprompt