creating-mcp-servers

Pass

Audited by Gen Agent Trust Hub on Oct 13, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the agent to fetch technical documentation from gofastmcp.com using the web_fetch tool. This external site is used as the authoritative source for the FastMCP framework documentation.
  • [REMOTE_CODE_EXECUTION]: The skill includes a pattern in references/MANDATORY_PATTERNS.md for installing the uv package manager via a remote script (curl | sh) from astral.sh. This source is the official domain of a well-known software company, Astral.
  • [COMMAND_EXECUTION]: The skill instructions in SKILL.md require the execution of the local script scripts/create_mcpb.py and standard system commands like zip to package server files into bundles.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external data fetched via web_fetch from gofastmcp.com to guide server construction.
  • Ingestion points: Documentation URLs indexed in references/LLMS_TXT.md.
  • Boundary markers: None; the skill does not specify delimiters for external content.
  • Capability inventory: File system writes (bundling), execution of scripts/create_mcpb.py, and suggested patterns for subprocess.run and sqlite3 execution.
  • Sanitization: None; fetched content is applied directly to implementation phases.
  • [DYNAMIC_EXECUTION]: The skill provides implementation templates in references/GATEWAY_PATTERNS.md that utilize dynamic execution techniques.
  • The 'Skills Gateway' pattern uses subprocess.run to execute Python scripts found in local directories such as ~/.claude/skills.
  • The 'Query Gateway' pattern suggests a database interface using f-string interpolation for SQL queries, which is a vulnerable pattern compared to standard parameterized queries.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 13, 2026, 07:04 PM