The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes external binaries ffmpeg and ffprobe via the subprocess module in scripts/sprite_cycles.py. These calls are used to extract technical metadata and raw frames from video clips generated during the asset creation process. The commands are executed using argument lists with no shell invocation, which is a secure practice for local tool usage.
[INDIRECT_PROMPT_INJECTION]: The skill defines an attack surface for indirect prompt injection because it ingests user-supplied text to generate AI content that is subsequently processed by local scripts.
Ingestion points: User-provided character descriptions and environment settings are interpolated into templates located in references/prompt-templates.md.
Boundary markers: The prompt templates do not utilize specific delimiters or instructions to prevent the model from obeying instructions embedded within user input.
Capability inventory: The skill has the capability to execute shell commands (ffmpeg, ffprobe) and perform file system write operations through its scripts.
Sanitization: The scripts do not perform sanitization or validation of the user-supplied strings prior to their use in model prompts.
[DYNAMIC_EXECUTION]: The script scripts/side_sprites.py dynamically modifies the Python search path (sys.path.insert) to import a local helper module, sprite_cycles.py. This is a standard method for organizing multi-file script toolkits and does not involve executing untrusted or remote code.

scenario-side-view-game-kit