AGENT LAB: SKILLS

book-cover-design

Fail

Audited by Gen Agent Trust Hub on Feb 19, 2026

Risk Level: CRITICALREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • Remote Code Execution (CRITICAL): The skill contains the command curl -fsSL https://cli.inference.sh | sh. This pattern is highly dangerous as it downloads and executes a script from the internet without any integrity checks or human review.
  • External Downloads (MEDIUM): The skill uses npx skills add to fetch external logic from inference-sh/skills, which is not a trusted source according to the established security parameters.
  • Command Execution (HIGH): The skill is granted Bash(infsh *) permissions to run commands through a CLI that is installed via an unverified remote script.
  • Indirect Prompt Injection (LOW): The skill takes untrusted user input and interpolates it into bash command arguments for image generation, creating a surface for indirect prompt injection. * Ingestion points: User-defined prompts for image generation in SKILL.md. * Boundary markers: Absent. * Capability inventory: Access to the Bash(infsh *) tool. * Sanitization: Absent.
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 19, 2026, 03:41 AM