mcp-optimist

The core server performs static code analysis locally, with a stated design principle of 'Sandboxed Analysis: No code execution during analysis' and 'No Network Access: All analysis performed locally'. However, the utility scripts (`tools/ask_kimi.js`, `tools/ask_kimi_then_qwen.js`) utilize `child_process.execSync` and `spawn` to invoke internal tools and external LLMs (via the `consult_ollama` tool). While these calls are currently controlled to execute trusted local scripts/tools, dynamic code generation and execution (`fs.writeFileSync` then `node <temp_file>`) in `tools/ask_kimi_then_qwen.js` for LLM interaction is a potential vector for code injection if the LLM output is not properly sanitized before embedding it into executable code. No obvious hardcoded secrets or arbitrary network risks were found in the core server logic.