Server has well-structured tool definitions with complete input schemas and descriptions. All 6 tools are explicitly registered with proper Pydantic models. Tool names follow verb_noun convention (solve_*). Descriptions are present and substantive (100-200 chars typical). However, output schemas lack explicit documentation in the source code, return types are inferred from Pydantic models but not formally documented for LLM consumption. Parameter descriptions are generally good but some lack constraint details (e.g., max_chars range is documented but not all enum values are explained in descriptions). Error handling is minimal, no recovery guidance or actionable error messages visible in the tool implementations.

• Document output schemas explicitly in tool descriptions or via a schema field. For each tool, state: 'Returns a JSON object with fields: {field_name: type, ...}'. Example: 'Returns {status: string, work_item_id: string, session_id: string, created: boolean, message: string, artifacts: object}'.
• Add try-catch blocks around runtime_run_solve_round() and state management calls. Return structured error responses with actionable guidance. Example: 'Failed to run solve round: {error_type: "RuntimeNotImplementedError", message: "Antigravity solver not configured", recovery: "Ensure Antigravity API credentials are set in environment variables."}'.
• Expand parameter descriptions to explain enum values and constraints. For work_item_mode, add: 'auto: let the system decide; continue-likely: prefer resuming existing work; new-likely: prefer creating new work; continue-explicit: must resume; new-explicit: must create new.'
• Add confirmation/dry-run support for destructive operations. solve_new_session and solve_run_round modify state, consider adding a 'dry_run' parameter or a separate 'preview_solve_round' tool to let agents validate before committing.
• Document result limits and pagination. For solve_get_latest_artifacts, state: 'Returns up to 10 artifacts per category (code, logs, decisions). Use session_id to iterate across sessions.' For solve_get_status, clarify what 'round' represents and how to fetch historical rounds.