Server exposes 5 tools with basic schemas and descriptions. All tools have input schemas with type definitions and descriptions present. However, descriptions are minimal (10-50 chars) and lack LLM-optimized guidance on WHEN to use each tool or what happens next. Parameter descriptions are present but terse. No output schemas documented. Error handling is minimal, tools return raw exceptions rather than actionable recovery guidance. Security-critical tool (run_command) lacks confirmation/dry-run pattern despite IRREVERSIBLE risk classification. No tool annotations (readOnlyHint, destructiveHint, idempotentHint) despite clear risk levels.

• Add tool annotations: mark run_command with destructiveHint=true, read_file/list_dir with readOnlyHint=true, write_file with destructiveHint=true. This signals safety properties to the LLM.
• Expand descriptions to 100-200 chars. E.g., run_command: 'Execute a shell command in WORKSPACE_ROOT. WARNING: irreversible and can access any file the server user can reach. Use for system administration tasks like installing packages, restarting services, or running scripts. Returns stdout and stderr.'
• Document output schemas for each tool. E.g., read_file returns {content: string, size_bytes: int, encoding: string}; system_metrics returns {disk: {used_gb, total_gb, percent}, memory: {used_mb, total_mb, percent}, top_processes: [{pid, name, cpu_percent, memory_mb}]}.
• Implement confirmation pattern for run_command: add optional 'dry_run' parameter (default false) that shows the command without executing it. Or require explicit 'confirm=true' for commands matching dangerous patterns (rm, dd, mkfs, etc.).
• Enhance error messages with recovery hints. E.g., PermissionError: 'Path /etc/passwd is outside workspace /home/user. Allowed paths: /home/user and subdirectories. Try list_dir(".") to see available files.' This guides the LLM to self-correct.

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

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