The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-provided video briefs during the 'Brief Intake' step in references/pre-production.md.
Ingestion points: Data enters the context via user-provided briefs (topic, duration, platform, brand, style) in references/pre-production.md.
Boundary markers: The skill lacks explicit boundary markers or instructions to the model to ignore embedded malicious instructions within the brief data.
Capability inventory: The skill possesses capabilities to execute shell commands via ffmpeg, ffprobe, egaki CLI, and curl for API interactions, as detailed in references/production.md and references/validation-export.md.
Sanitization: There is no evidence of explicit sanitization or escaping of the user-provided 'visual_description' or 'text_overlay' before they are interpolated into CLI commands.
[COMMAND_EXECUTION]: The skill orchestrates multiple CLI tools to produce video content. It utilizes ffmpeg for concatenation, audio mixing, and subtitle burn-in, ffprobe for codec validation, and egaki CLI for interfacing with AI video generation models (references/production.md, references/validation-export.md).
[DYNAMIC_EXECUTION]: The skill generates temporary artifacts at runtime, including .srt subtitle files, shell commands for video processing, and a video-manifest.json export log (references/production.md, references/validation-export.md). These are generated from templates using inputs derived from the brief and brand guidelines.

paw-cra-video-shortform