Claude SDK Expert
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDYNAMIC_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [COMMAND_EXECUTION]: The skill documents the capability for agents to execute shell commands via a Bash tool. It includes code examples demonstrating how to sanitize input by blocking dangerous patterns such as fork bombs (':(){ :|:& };:') and destructive commands ('rm -rf').
- [REMOTE_CODE_EXECUTION]: The documentation shows how to integrate with Model Context Protocol (MCP) servers using 'npx' and 'docker' to dynamically pull and execute tool configurations from the Model Context Protocol ecosystem.
- [DYNAMIC_EXECUTION]: The skill describes architecture patterns where agents can autonomously create, modify, and run scripts to perform iterative debugging and multi-step tasks.
- [DATA_EXFILTRATION]: Implementation examples include the use of web-access tools (WebFetch, WebSearch) which allow agents to make network requests to external domains for research and data gathering.
Audit Metadata