search
Pass
Audited by Gen Agent Trust Hub on Oct 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from search engines and web pages, which could contain malicious instructions designed to influence the agent's behavior.
- Ingestion points: Results from
bdata searchandbdata discovercommands, as well as full page content when the--include-contentflag is used (detailed inSKILL.mdandreferences/examples.md). - Boundary markers: While the skill expects structured JSON output and uses
jqfor parsing, it lacks explicit instructions for the agent to ignore or isolate instructions potentially embedded within the retrieved search results or page bodies. - Capability inventory: The skill utilizes file system writing (
-oflag), network communication (via thebdataCLI andcurl), and shell command execution (xargswithbash -c). - Sanitization: A legacy
curlfallback example inreferences/patterns.mdusesjq -sRr @urito sanitize search queries, but there is no evidence of sanitization for the content retrieved from external pages before it is processed by the agent. - [COMMAND_EXECUTION]: The skill relies on the
bdataCLI tool and various shell utilities to execute its functions, involving complex command pipelines. - Evidence: Batch processing patterns in
references/patterns.mdusexargscombined withbash -cto execute scraping tasks for multiple URLs in parallel. - The instructions provide patterns for shell loops that handle user-provided queries or data from external files (
queries.txt), which are then passed as arguments to CLI commands.
Audit Metadata