Strong tool definitions with complete schemas and detailed descriptions. All 9 tools have explicit input schemas with proper JSON Schema structure (type, properties, required fields, additionalProperties=false). Descriptions are comprehensive (100-300+ chars), explaining WHAT the tool does, WHEN to use it, and critical safety constraints. Tool names follow verb_noun pattern (xshell_*, sftp_*). However, output schemas are not documented, LLMs cannot predict response structure. Parameter descriptions are thorough but some lack explicit format/constraint details. Error handling guidance is implicit in descriptions rather than explicit recovery patterns. Security-conscious design (user approval gates, credential handling) is well-documented but not formalized as error categories.

• Document output schemas for all tools. Example for xshell_list_sessions: 'Returns {sessions: [{session_id: string, metadata: {rows: number, columns: number, ...}}], total: number}'.
• Add explicit error classification to tool descriptions. Example: 'On user rejection: returns {rejected: true, reason: string}, agent should ask user for clarification. On timeout: returns {timeout: true}, agent may retry. On invalid session_id: returns {error: "session not found", available_sessions: [...]}'.
• Clarify session_id parameter across all tools: 'Session ID from xshell_list_sessions.sessions[].session_id. If only one session is online, this parameter is optional.'
• Document sftp_download and sftp_upload return values and the expected workflow with sftp_transfer_status. Example: 'Returns {transfer_id: string}. Poll sftp_transfer_status(transfer_id) to track progress through stages: authentication_and_transfer → sha256_verification → completed.'
• Add constraint descriptions to numeric parameters. Example for timeout_ms: 'Timeout in milliseconds (100 - 120000). Default 30000 (30 seconds). Longer timeouts for slow networks or large file transfers.'