The Agent Skills Directory

[SAFE]: The skill demonstrates best practices for the platform it describes, including sequential processing to avoid resource exhaustion (CUDA OOM) and quality assurance steps before expensive video rendering.
[COMMAND_EXECUTION]: The skill uses the infsh CLI tool to execute various AI modules (e.g., pruna/p-image, inworld/text-to-speech-2, infsh/media-merger). These are the primary tools for the skill's intended purpose.
[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow that ingests untrusted external data (text for scripts and image URLs for training), which is then processed by external AI models. This is an inherent surface for indirect prompt injection, although no malicious exploitation is present in the skill itself.
Ingestion points: User-provided script content in inworld/text-to-speech-2 and image source URLs in phota/train (SKILL.md).
Boundary markers: No specific delimiters or "ignore previous instructions" warnings are used in the prompt examples.
Capability inventory: The skill uses Bash, Read, Write, and Agent tools to manage files and invoke external AI applications.
Sanitization: The skill does not specify manual sanitization of the input text before processing.

ai-podcast