mcp-code-execution

Pass

Audited by Gen Agent Trust Hub on Oct 26, 2026

Risk Level: SAFEDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [DYNAMIC_EXECUTION]: The skill provides an operating procedure for the agent to generate and execute TypeScript code to interface with Model Context Protocol (MCP) servers. It explicitly mandates a secure sandbox, resource limits, and monitoring as a requirement for this workflow.
  • [INDIRECT_PROMPT_INJECTION]: The proposed architecture involves the agent reading and writing files in a workspace (e.g., ./workspace/, ./servers/, ./skills/) where tool-generated content is stored. This creates an ingestion surface for untrusted data. The skill suggests boundary markers (PII tokenization) and sanitization (filtering results) to mitigate risk. Capabilities include file operations, tool wrapper calls, and agent-generated script execution.
  • [EXTERNAL_DOWNLOADS]: The skill references official engineering documentation from Anthropic to provide technical context for the token-saving methodology.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 26, 2026, 08:16 AM