The Agent Skills Directory

[PROMPT_INJECTION]: The skill creates a surface for indirect prompt injection by instructing the agent to ingest untrusted data (transcripts) and use that data to generate shell commands (FFmpeg) or edit decision lists.
Ingestion points: Video transcripts and recording metadata in Layer 2 (Organization).
Boundary markers: Absent; example prompts do not include delimiters or instructions to ignore embedded commands in the transcript.
Capability inventory: High-privilege file manipulation via ffmpeg and external asset rendering via npx remotion.
Sanitization: Absent; the skill does not suggest validating or escaping content from the transcripts before command generation.
[COMMAND_EXECUTION]: The skill provides numerous templates for shell-based operations, including complex FFmpeg pipelines for cutting, scaling, and scene detection, as well as Node.js execution through npx remotion. While standard for video editing, these provide a significant capability that could be misdirected by malicious input.
[EXTERNAL_DOWNLOADS]: The skill uses requests to communicate with the official ElevenLabs API for speech generation. This is a well-known service and the implementation correctly utilizes environment variables (ELEVENLABS_API_KEY) for secure credential management.

video-editing