Single tool 'delegar_a_opencode' has critical definition gaps. Input schema visible but severely underdeveloped: only 'instruction' parameter with minimal description. No output schema documented. Tool description is Spanish-language and vague ('Delega una tarea...'). No parameter constraints, enums, or validation guidance. No error handling patterns. Tool performs WRITE operations (code execution + security audit) but lacks idempotency markers, confirmation patterns, or permission gates.

• Add explicit output schema documenting return type (e.g. {result: string, execution_time_ms: number, audit_decision: 'approved'|'rejected', audit_feedback: string}).
• Constrain 'instruction' parameter: add maxLength, minLength, and description explaining expected format (e.g. 'Natural language task description, 10-2000 chars, no shell metacharacters').
• Implement confirmation pattern: add optional 'confirm' boolean parameter or separate 'confirm_task_execution' tool to gate destructive operations.
• Document error responses with recovery guidance: 'Audit rejected. Suggestion: {suggestion}. Retry with modified instruction or escalate to human.' instead of silent approval.
• Rewrite description in English, 50-200 chars, answering: What does it do? When use it? What does it return? E.g. 'Execute a programming task via OpenCode with mandatory security audit by Hermes. Returns execution result and audit decision.'