The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions frequently generate Bash commands using user-provided input, such as filenames (e.g., gencast <input-file>). There is a risk of shell injection if the agent interpolates filenames containing shell metacharacters (e.g., ;, &, |) without proper sanitization.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection. It ingests untrusted external data (Markdown and PDF documents) to generate podcast scripts and audio.
Ingestion points: Files provided by the user as input to the gencast CLI (e.g., document.md, chapter.md).
Boundary markers: None identified; document content is processed directly by the external tool.
Capability inventory: Uses the Bash tool to run commands and the Read tool to access files.
Sanitization: There is no evidence of sanitization or content filtering for the documents before they are processed by the tool's LLM components.
[EXTERNAL_DOWNLOADS]: The skill suggests installing a third-party package (pip install gencast) if it is not present. While this is a standard developer workflow, users should verify the package's integrity before installation.

gencast