The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to invoke ffmpeg and ffprobe as subprocesses for scaling images, generating Ken Burns effects, and muxing audio with video. The commands include the -y flag, which forces the overwriting of output files without user confirmation.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process user-supplied story beats, narration text, and visual descriptions, creating an attack surface for indirect prompt injection.
Ingestion points: External data enters the workflow via user-defined titles, emotional arcs, and beat-specific prompts (e.g., edit_prompt, narration, video_prompt) which are stored in plan.json and media filenames.
Boundary markers: The workflow incorporates human-in-the-loop checkpoints, requiring explicit phrases like 'approve plan', 'approve stills', 'approve audio', and 'approve clips' before proceeding to high-cost or final assembly phases.
Capability inventory: The agent has the ability to write to the local filesystem, perform network requests to Pruna and Replicate APIs, and execute shell commands via media processing tools.
Sanitization: There is no evidence of automated sanitization or escaping for user-provided strings before they are interpolated into prompt templates or command-line arguments, relying entirely on user review during the approval gates.
[EXTERNAL_DOWNLOADS]: The skill identifies several prerequisite skills from the PrunaAI organization that must be installed via npx to provide core image and video generation capabilities. These represent vendor-owned dependencies necessary for the skill's primary function.

illustrated-story-reel