Five tools with complete schemas and descriptions, but significant quality gaps. Tool names lack clear action verbs (cavalry_ping, cavalry_run_script are acceptable; cavalry_search_knowledge and cavalry_get_scene_info are weak). Descriptions are present but generic, most under 100 chars, lacking WHEN/WHY guidance. Parameters mostly undescribed (e.g., 'script' in cavalry_run_script has description, but 'filePath' in cavalry_save_scene lacks detail on format/behavior). Output schemas are not documented, tools return text blobs or write to files rather than structured JSON. Error handling is minimal: tools return 'Success' or error strings without recovery guidance. No tool annotations (readOnlyHint, destructiveHint). cavalry_run_script is particularly risky, executes arbitrary code with no validation or dry-run option.

• Add tool annotations: mark cavalry_run_script with destructiveHint=true, cavalry_ping with readOnlyHint=true, cavalry_save_scene with destructiveHint=true. This signals to agents which tools have side effects.
• Implement a dry-run or confirmation step for cavalry_run_script. Return a preview of what the script will do, then require explicit confirmation before execution. Prevents accidental data loss.
• Document output schemas for all tools. cavalry_search_knowledge should return {results: [{id, source, content, relevance_score}], total_count}. cavalry_get_scene_info should return {layers: [{id, name, type, properties}], composition_info: {...}}.
• Enhance parameter descriptions with format/constraint details. cavalry_save_scene filePath: 'Optional absolute or relative file path. If omitted, overwrites current scene. Must end in .cavalry.' cavalry_run_script script: 'Cavalry JavaScript code (max 10KB). Must use api.* methods; direct DOM access forbidden.'
• Improve error messages with recovery guidance. Instead of 'Error: script failed', return 'Script execution failed: [specific error]. Check Cavalry console for details. Try cavalry_ping first to verify Stallion bridge is reachable.'