ai-social-media-content

Fail

Audited by Gen Agent Trust Hub on Mar 4, 2026

Risk Level: HIGHREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [REMOTE_CODE_EXECUTION]: The skill instructs the agent or user to install a tool using the command curl -fsSL https://cli.inference.sh | sh. This is a high-risk pattern that executes a remote script from an untrusted domain without user inspection, allowing for arbitrary code execution on the host system.
  • [COMMAND_EXECUTION]: The skill requires the Bash(infsh *) capability. It uses this to execute complex shell commands, including for-loops to batch process content and file redirection (>) to store JSON outputs locally. While aligned with the skill's purpose, it provides a significant attack surface if the input data is compromised.
  • [EXTERNAL_DOWNLOADS]: The installation process fetches pre-compiled binaries from dist.inference.sh. Although the documentation claims checksum verification is performed, the integrity of the initial installation script (cli.inference.sh) remains a single point of failure.
  • [INDIRECT_PROMPT_INJECTION]: The skill demonstrates a multi-step workflow where the output from one AI model (e.g., Claude) is saved to a file (script.json) and then used as input for another tool (infsh app run infsh/kokoro-tts).
  • Ingestion points: Untrusted data from AI model responses is saved to script.json and voice.json (SKILL.md).
  • Boundary markers: None present; the content is interpolated directly into subsequent command inputs.
  • Capability inventory: Subprocess calls via infsh, file-write operations via Bash redirection.
  • Sanitization: No evidence of escaping or validation of the content before it is passed to the next stage of the pipeline.
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
Risk Level
HIGH
Analyzed
Mar 4, 2026, 12:29 PM