ai-rag-pipeline

Fail

Audited by Gen Agent Trust Hub on Feb 17, 2026

Risk Level: CRITICALEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • EXTERNAL_DOWNLOADS / REMOTE_CODE_EXECUTION (CRITICAL): The skill documentation and examples explicitly use the pattern curl -fsSL https://cli.inference.sh | sh. This executes a remote script with shell privileges without verification. The source inference.sh is not a designated trusted source.
  • COMMAND_EXECUTION (HIGH): The skill requests Bash(infsh *) capability. This allows the agent to execute any sub-command of the infsh tool, which includes running arbitrary applications, logging in, and potentially modifying local environment configurations.
  • PROMPT_INJECTION (LOW): Indirect Prompt Injection Surface. The skill is designed to fetch content from the live web and URLs and then feed that content directly into LLM prompts.
  • Ingestion points: Data enters from tavily/search-assistant, exa/search, and tavily/extract tools.
  • Boundary markers: Absent. The skill uses direct shell variable interpolation (e.g., $SEARCH_RESULT) inside JSON strings without delimiters or instructions for the model to ignore embedded commands.
  • Capability inventory: The infsh tool can execute remote apps and interact with external APIs.
  • Sanitization: None observed. The external content is treated as trusted context.
  • DATA_EXFILTRATION (LOW): The skill sends research data and extracted URL content to the inference.sh platform and third-party LLM providers (OpenRouter). While this is the intended function, it involves moving potentially sensitive research data to external services.
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, 06:42 PM