ai-podcast-creation
Fail
Audited by Gen Agent Trust Hub on Feb 18, 2026
Risk Level: CRITICALREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- REMOTE_CODE_EXECUTION (CRITICAL): The skill includes an installation command
curl -fsSL https://cli.inference.sh | sh. This 'curl pipe to sh' pattern executes unverified code from an untrusted external domain directly on the user's system, providing a direct path for complete system compromise. - COMMAND_EXECUTION (HIGH): The skill is configured with
allowed-tools: Bash(infsh *)and provides numerous examples for running shell commands. This capability, combined with the untrusted CLI, poses a high risk of arbitrary command execution. - EXTERNAL_DOWNLOADS (MEDIUM): The skill attempts to download the
infshCLI tool and several external agent skills (inference-sh/skills@...) from a source that is not included in the pre-approved trusted organizations list. - PROMPT_INJECTION (LOW): The skill exhibits an Indirect Prompt Injection surface (Category 8). It processes untrusted document content and interpolates it into prompts for the Claude LLM and various TTS models without using sanitization or boundary markers.
- Ingestion points:
SKILL.md(via<your-document-content>and script generation examples). - Boundary markers: Absent; untrusted content is directly embedded in JSON strings.
- Capability inventory:
infshsubprocess calls, network access to OpenRouter and Inference.sh APIs. - Sanitization: None detected in the provided workflows.
Recommendations
- HIGH: Downloads and executes remote code from: https://cli.inference.sh - DO NOT USE without thorough review
- AI detected serious security threats
Audit Metadata