prompt-engineering

Fail

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: CRITICALREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
  • Remote Code Execution (CRITICAL): The skill instructs the agent to run the command curl -fsSL https://cli.inference.sh | sh. This is a classic 'curl pipe sh' vulnerability that allows an untrusted remote domain to execute arbitrary code with shell privileges on the local system.
  • External Downloads (HIGH): The 'Related Skills' section encourages the use of npx skills add to download and install additional content from an untrusted source (inference-sh/skills). This can lead to the silent installation of malicious scripts or tool configurations.
  • Indirect Prompt Injection (LOW): The skill provides templates for processing untrusted data (e.g., [article text], [code]) without using delimiters or sanitization.
  • Ingestion points: External data is interpolated into the --input flag of infsh commands in SKILL.md.
  • Boundary markers: Absent; no instructions are provided to the model to ignore instructions embedded in the user-provided data.
  • Capability inventory: The skill utilizes the Bash(infsh *) tool to interact with various AI models.
  • Sanitization: No escaping or validation of external content is performed before passing it to the model.
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
CRITICAL
Analyzed
Feb 17, 2026, 06:27 PM