bx-search
Fail
Audited by Gen Agent Trust Hub on Jul 2, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Fetches the
bxbinary and installation scripts from the Brave Search CLI repository on GitHub (github.com/brave/brave-search-cli). - [REMOTE_CODE_EXECUTION]: Provides installation commands that download and execute remote scripts from the Brave GitHub repository using the
shandiex(Invoke-Expression) interpreters. - [COMMAND_EXECUTION]: Instructs the agent to perform web search operations and manage settings by executing the
bxbinary via shell commands. - [PROMPT_INJECTION]: The skill processes untrusted web content from search results, which creates a surface for indirect prompt injection.
- Ingestion points: Data is ingested into the agent context through results from the
bx context,bx answers, andbx webcommands (SKILL.md). - Boundary markers: No specific delimiters or instructions are used to signal that search output contains potentially untrusted content.
- Capability inventory: The skill requires the ability to execute shell commands (SKILL.md).
- Sanitization: There is no explicit sanitization of the search results before they are provided to the agent.
Recommendations
- HIGH: Downloads and executes remote code from: https://raw.githubusercontent.com/brave/brave-search-cli/main/scripts/install.sh - DO NOT USE without thorough review
Audit Metadata