The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently executes the renku CLI and local Node.js scripts (validate-image-prompt-guides.mjs, validate-video-prompt-guides.mjs) to manage project state and validate configurations. These commands are executed locally using child process modules.
[PROMPT_INJECTION]: The skill contains an indirect prompt injection surface (Category 8). It ingests narrative and project context from local files and CLI outputs, which are then interpolated into prompts for external image and video generation tools.
Ingestion points: Untrusted data from Scene Beats and Shot Plans enters the agent context via renku generation context and renku screenplay beats show (referenced in references/scene-storyboard-sheet.md and SKILL.md).
Boundary markers: There are no explicit delimiters or specific instructions for the AI to ignore malicious commands embedded within the narrative text segments of the prompt (e.g., facts.contextText).
Capability inventory: The skill allows the agent to trigger paid generation tasks via renku generation run and perform built-in Codex image generation.
Sanitization: The skill does not implement sanitization or escaping of narrative context before it is included in generation prompts.

media-producer