Three tools with Russian descriptions and basic schemas. All tools have input schemas with type definitions, but descriptions are minimal (10-50 chars), lack actionable context, and omit prerequisites/side effects. No output schemas documented. Error handling is generic (timeout/exception strings). Tool names lack clear verb-noun patterns. Parameters lack constraints (enums, ranges). No guidance on when to use each tool vs. alternatives. Composition is weak: bash_run is dangerously generic; antigravity_run and bash_run overlap in capability.

• Expand tool descriptions to 100-200 chars. For antigravity_run: 'Execute Antigravity CLI agent with a task prompt. Use this for AI-powered task automation. Requires agy CLI installed. Returns agent output or error. Use bash_run for direct shell commands instead.'
• Add output schema documentation. E.g., antigravity_run returns {"output": string, "status": "success"|"timeout"|"error", "duration_seconds": number}. bash_run returns {"stdout": string, "stderr": string, "exit_code": number}.
• Implement input validation and sanitization for bash_run. Reject dangerous patterns (rm -rf, >, |, &, ;). Return actionable error: 'Command contains pipe operator. Use bash_run for simple commands only; chain via separate tool calls.'
• Add confirmation step for destructive bash_run commands. Detect patterns like 'rm', 'delete', 'truncate' and return {"requires_confirmation": true, "preview": "Would delete /path/to/file"}. Require explicit confirm_destructive_command call.
• Rename tools for clarity: antigravity_run → execute_antigravity_task; bash_run → execute_shell_command; system_status → get_system_status. Add parameter descriptions: prompt → 'Task description for Antigravity agent (max 1000 chars)', command → 'Shell command to execute (no pipes/redirects; use separate calls)'.