ai-rag-pipeline

Fail

Audited by Gen Agent Trust Hub on Mar 8, 2026

Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: Fetches the CLI installation script and binary from the vendor's official domains (cli.inference.sh and dist.inference.sh).
  • [REMOTE_CODE_EXECUTION]: The Quick Start guide recommends an installation method using curl -fsSL https://cli.inference.sh | sh, which pipes a remote script directly into the shell for execution.
  • [COMMAND_EXECUTION]: The skill requires the Bash(infsh *) tool permission to execute the vendor's CLI commands for searching and running LLM models.
  • [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests untrusted data from external web sources and interpolates it into prompts without sufficient sanitization or boundary markers.
  • Ingestion points: Web search results from Tavily and Exa, as well as content extracted from arbitrary URLs.
  • Boundary markers: Absent; untrusted variables like $SEARCH, $CONTENT, and $EVIDENCE are placed directly into the LLM prompt strings.
  • Capability inventory: The skill can execute arbitrary subcommands of the infsh CLI tool.
  • Sanitization: No escaping or validation is performed on the retrieved web content before it is sent to the LLM.
Recommendations
  • HIGH: Downloads and executes remote code from: https://cli.inference.sh - DO NOT USE without thorough review
Audit Metadata
Risk Level
HIGH
Analyzed
Mar 8, 2026, 02:51 AM