mcp-server-builder
Pass
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external OpenAPI specifications to generate tool definitions, creating an attack surface for indirect prompt injection.
- Ingestion points:
scripts/openapi_converter.pyreads and parses OpenAPI JSON/YAML files from the local filesystem (e.g.,assets/sample_openapi.json). - Boundary markers: No explicit delimiters or instructions are used in the converter or scaffolder to isolate untrusted descriptions from the agent's system prompt.
- Capability inventory: The resulting tools generated by this skill and the examples provided emphasize shell-based operations, file system access, and command execution (e.g.,
ripgrep,pnpm). - Sanitization: The converter sanitizes tool names using
sanitize_namebut passes summary and description text directly into the generated JSON definitions without content validation. - [DYNAMIC_EXECUTION]: The script
scripts/server_scaffolder.pyis capable of dynamically generating and writing TypeScript or Python source code (e.g.,src/index.ts,server.py) to an output directory based on provided tool configurations. This capability assembles executable boilerplate at runtime. - [COMMAND_EXECUTION]: The provided reference documentation in
references/server-implementation.mdcontains code examples for tool handlers usingchild_process.execSyncandsubprocess.run. These templates, while intended as developer guides, include shell execution patterns (specifically theripgrepregex search) that could be vulnerable to command injection if implemented by a user without robust input sanitization beyond the simple single-quote replacement shown.
Audit Metadata