The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructs the agent to read docs/TOOLS-AND-API-KEYS.md to check for available credentials and service setup. Accessing files explicitly named to contain sensitive API keys and authentication tokens increases the risk of data exposure or accidental exfiltration.
[COMMAND_EXECUTION]: The workflow requires the execution of external multimedia tools and local scripts to process video assets, including:
ffprobe for inspecting duration, streams, and resolution.
node scripts/transcribe-elevenlabs.mjs for transcription services.
[INDIRECT_PROMPT_INJECTION]: The skill processes raw talking-head videos and transcripts which can be sourced from external providers (e.g., ElevenLabs, OpenAI Whisper). It lacks explicit boundary markers or sanitization to prevent the agent from executing instructions that might be embedded in the transcript text.
Ingestion points: Video recordings, source recordings in video-projects/<slug>/, and transcript files used for visual beat sheets.
Boundary markers: Absent; the agent is instructed to read transcripts directly into context for mistaking cutting and visual storytelling.
Capability inventory: File system read/write access, shell execution for media tools (ffprobe), and local Node.js script execution.
Sanitization: No validation or filtering of transcript content is mentioned before it is used to anchor sub-compositions and visual beats.

edit-video