The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes word-level transcripts provided as JSON files, which represents a potential attack surface if the transcript contains malicious instructions.
Ingestion points: scripts/find-cut-candidates.mjs and scripts/apply-cuts.mjs read external transcript files.
Boundary markers: The skill instructions in SKILL.md define a "review-gated" workflow, requiring the agent and user to approve proposed cuts, which serves as a human-in-the-loop safety measure.
Capability inventory: The skill can execute shell commands (ffmpeg, ffprobe) and perform file system operations (write/read).
Sanitization: The scripts implement strict validation for cut ranges, ensuring they are finite numbers and within duration bounds before incorporating them into ffmpeg filtergraphs.
[COMMAND_EXECUTION]: The skill uses spawnSync to execute ffmpeg and ffprobe. It correctly passes arguments as an array to prevent shell injection. It also uses npx serve to launch a local development server for reviewing edits.

cut-mistakes