The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data which creates an indirect prompt injection surface.\n
Ingestion points: Raw video files (SRC), reference video URLs, and B-roll folders.\n
Boundary markers: Absent; the skill lacks explicit instructions to ignore embedded commands in media content.\n
Capability inventory: Extensive file system access, execution of ffmpeg with complex filter graphs, and network access via yt-dlp.\n
Sanitization: File integrity is validated via SHA-256, but the content being transcribed and analyzed is not sanitized for malicious instructions.\n- [DYNAMIC_EXECUTION]: The skill utilizes awk to dynamically assemble ffmpeg filter graph scripts at runtime based on segment timings. This programmatic generation of command logic adds complexity to the execution environment. Found in ffmpeg-recipes.md.\n- [REMOTE_CODE_EXECUTION]: The skill invokes npx -y hyperframes to download and execute the HyperFrames CLI from the NPM registry. It also uses yt-dlp to download content from user-provided URLs for analysis. Found in SKILL.md and style-decode.md.\n- [EXTERNAL_DOWNLOADS]: The skill downloads reference videos, B-roll assets, and music from external sources such as YouTube, Pexels, and Pixabay. These downloads are documented for attribution but originate from external infrastructure.

raw-video-editor