KGB Agent exposes 3 process management tools with CRITICAL gaps. Tool descriptions are present but minimal (10-50 chars, well below the 194-char baseline). Input schemas are visible but lack parameter descriptions entirely, the 'command' param in run_background_process has no guidance on format or constraints. No output schemas documented. Error handling is absent, no recovery guidance, no categorization of retryable vs fatal errors. The tool names are verb-based (run_, stop_, find_) but lack clarity on scope and side effects. No evidence of idempotency, permission gates, or audit trails for destructive operations (stop_background_process). STDIO transport caps overall quality.

• Add 50 - 150 character descriptions to each tool explaining WHAT it does, WHEN to use it, and what it returns. Example: 'Start a background process and return its PID. Use this to launch long-running tasks (servers, workers). Returns {pid, status}. Note: process runs in the specified cwd and is tracked by this server.'
• Document parameter descriptions for all inputs. For 'command', specify: 'Array of command arguments (e.g., ["python", "script.py", "--verbose"]). Do not use shell syntax; pass arguments as separate list items.'
• Define output schemas for each tool. Example for run_background_process: {pid: integer, status: string, started_at: ISO8601 timestamp}. For find_process_by_port: {pid: integer, command: string, port: integer}.
• Add error handling with recovery guidance. Example: 'If PID not found, return {error: "Process not found", available_pids: [123, 456]}' to enable agent self-correction.
• For stop_background_process, add a dry_run parameter (boolean, default false) to preview what will be killed before executing. Prevents accidental termination of critical processes.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry