ai-podcast-creation
Pass
Audited by Gen Agent Trust Hub on Mar 25, 2026
Risk Level: SAFEREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill instructs users to install the
infshcommand-line tool by piping a script fromhttps://cli.inference.shdirectly to the shell (sh). This pattern executes remote code on the local system. The script originates from the vendor's own infrastructure and is the documented method for installation. - [EXTERNAL_DOWNLOADS]: The installation script downloads binaries and checksums from
dist.inference.shto provide the CLI functionality. - [COMMAND_EXECUTION]: The skill uses the
Bashtool to executeinfshcommands, which orchestrate various AI tasks like text-to-speech conversion, music generation, and audio merging. - [PROMPT_INJECTION]: The skill demonstrates processing external data (e.g.,
<your-document-content>) by passing it into LLM prompts and TTS models, creating an indirect injection surface. - Ingestion points: Data is ingested via command-line arguments in
SKILL.mdexamples. - Boundary markers: No explicit delimiters or 'ignore' instructions are used in the prompt examples to separate untrusted data from agent instructions.
- Capability inventory: The skill is restricted to the
Bash(infsh *)tool set. - Sanitization: No input sanitization or filtering is visible in the provided workflows.
Audit Metadata