The Agent Skills Directory

[DATA_EXFILTRATION]: The skill automates the discovery and upload of user media files to external cloud storage services.
Evidence: It uses the command find ~/Downloads ~/Desktop -maxdepth 1 -type f \( -iname "*.mp4" -o -iname "*.mov" -o -iname "*.webm" \) -mmin -15 to locate recently saved media files.
Evidence: Discovered files are transmitted to a remote S3 bucket via curl -X PUT using presigned URLs generated by vendor-specific tools.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted media content and uses the resulting analysis to drive subsequent video generation tools.
Ingestion points: Video and media analysis output generated by the arcads_analyze_media tool (SKILL.md Step 2).
Boundary markers: The instructions lack specific delimiters or safety warnings to the model to ignore potential malicious instructions embedded within the analyzed media text.
Capability inventory: Local file system access (find, mkdir), network operations (curl), and powerful media generation capabilities (Seedance 2.0).
Sanitization: No explicit content sanitization or validation is performed on the text generated from media analysis before it is interpolated into generation prompts.
[COMMAND_EXECUTION]: The skill utilizes shell commands for environment setup and data management.
Evidence: Shell execution includes mkdir -p for session organization, find for asset discovery, and curl for both uploading assets and downloading generated results.

clone-hook