The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions to download and execute a remote script directly using a pipe-to-shell pattern.
Evidence: curl -fsSL https://cli.inference.sh | sh in SKILL.md.
Context: This is used to install the infsh CLI tool, but executing unverified scripts from the internet is a major security risk.
[EXTERNAL_DOWNLOADS]: The skill downloads binary executables from a remote domain during the installation process.
Evidence: The install script fetches files from dist.inference.sh as mentioned in the documentation.
[COMMAND_EXECUTION]: The skill relies on the Bash tool to execute system commands.
Evidence: Extensive use of infsh app run commands throughout the instructions.
Mitigation: The skill correctly uses allowed-tools: Bash(infsh *) to limit the scope of shell execution to a specific CLI tool.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted external data which is then used as input for further AI processing.
Ingestion points: The 'NotebookLM-Style Content' section specifically instructs the agent to read document content (<your-document-content>) and generate scripts from it.
Boundary markers: Absent. Untrusted content is interpolated directly into the JSON input for the infsh tool without clear delimiters or 'ignore' instructions.
Capability inventory: The skill can invoke LLMs (Claude via OpenRouter), generate audio (TTS), and perform media merging via the infsh CLI.
Sanitization: No sanitization or validation of the input text is performed before it is passed to the generation tools.

ai-podcast-creation