Single tool 'run_command' has a minimal description (18 chars, below 20-char threshold) and a bare-bones schema with only a 'command' string parameter. No output schema documented. No error handling guidance, no parameter constraints, no security warnings despite the tool being DESTRUCTIVE. The description 'Run a shell command' is too generic and does not explain when to use it, what it returns, or the irreversible consequences of execution. Parameter lacks format/constraint documentation. No idempotency, confirmation, or dry-run support for a destructive operation.

• Expand tool description to 50 - 150 chars: 'Execute a shell command on the server. Returns stdout/stderr. WARNING: Irreversible. Validate input to prevent injection. Use only for trusted, non-destructive commands.'
• Document output schema: {type: 'object', properties: {stdout: {type: 'string'}, stderr: {type: 'string'}, exitCode: {type: 'integer'}}, required: ['stdout', 'exitCode']}
• Add parameter constraints: command must be non-empty, max 1000 chars, no shell metacharacters (|, ;, &, $, backticks) without explicit escaping. Document in description.
• Implement confirmation pattern: add optional 'confirm' boolean param. If false/absent, return dry-run preview of what would execute. Require confirm=true for actual execution.
• Add error handling: return {error: 'Command failed', exitCode: N, stderr: '...', suggestion: 'Check syntax or permissions'} instead of raw exception. Guide LLM on recovery.