The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The Quick Start guide recommends installing the infsh CLI by fetching a remote script from https://cli.inference.sh and piping it directly to the shell (curl ... | sh). This execution method bypasses verification and could allow a compromised server to execute arbitrary code on the host machine.
[COMMAND_EXECUTION]: The skill uses the Bash tool to execute infsh commands. While restricted to the infsh prefix via frontmatter configuration, the recommended installation method involves executing unverified shell scripts, and the production workflows involve multiple shell commands for tool execution and session management.
[EXTERNAL_DOWNLOADS]: The skill downloads an external installer script from https://cli.inference.sh. As this domain is not within the defined trusted organization scope and was flagged as untrusted by automated scanners, fetching and executing its content constitutes a risk.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection in its audio production pipelines.
Ingestion points: The 'Full Episode Pipeline' and 'NotebookLM-Style Content' sections in SKILL.md ingest script content generated by the openrouter/claude-sonnet-45 model.
Boundary markers: Absent. The generated text is passed directly into the kokoro-tts tool using variable placeholders without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill possesses the Bash(infsh *) capability, which allows it to execute AI models and write files via redirection.
Sanitization: Absent. No escaping, validation, or filtering is performed on the output from the LLM before it is interpolated into subsequent tool commands.

ai-podcast-creation