The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection in the 'NotebookLM-Style Content' workflow where it processes untrusted documents.
Ingestion points: Raw document content from the user is interpolated directly into an LLM prompt in SKILL.md.
Boundary markers: The prompts lack clear delimiters or instructions to the agent to disregard potential commands hidden within the processed text.
Capability inventory: The skill utilizes the infsh tool via Bash to perform network-based API requests for audio generation and media merging.
Sanitization: No sanitization or validation of the input document is described or implemented.
[EXTERNAL_DOWNLOADS]: The skill references external resources for setup and extending functionality.
Provides links to installation instructions hosted on a GitHub repository (inference-sh/skills).
Instructs users on adding related components via the npx package manager from the vendor's repository.
[COMMAND_EXECUTION]: The skill's primary operations involve executing shell commands.
Uses the infsh CLI tool for inference and media processing.
Command execution is restricted to the infsh binary through the allowed-tools manifest configuration.

ai-podcast-creation