The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous templates for executing ffmpeg and ffprobe commands to process video files locally.
Evidence includes commands for audio extraction (ffmpeg -i raw.mp4 -vn -c:a aac), tonemapping (ffmpeg -i in.mp4 -vf zscale...), video chunking (ffmpeg -ss {in} -t {dur} -c:v libx264), and final assembly.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied scripts and tool-generated transcripts, which are interpolated into prompts for the generative video model.
Ingestion points: Raw video footage, user scripts, and word-level transcripts (from arcads_transcribe) are used as primary inputs in Phases A and C.
Boundary markers: Prompt templates use quotation marks for text overlays, but the skill lacks specific delimiters or "ignore previous instructions" warnings to prevent script content from influencing model behavior.
Capability inventory: The skill has access to file-system operations (via ffmpeg), network uploads, and generative AI tool calls (arcads_generate_video_omni_flash).
Sanitization: The instructions emphasize verification of spelling and timing but do not describe programmatic sanitization or escaping of input data before prompt construction.
[DYNAMIC_EXECUTION]: The skill generates shell command strings for ffmpeg from templates at runtime, including the creation of a local file list to facilitate video concatenation.
[DATA_EXFILTRATION]: The workflow involves uploading video and image assets to the vendor's storage infrastructure via arcads_get_upload_url and arcads_register_video.

omniflash-restyle