21 tools with complete schemas and descriptions, but significant quality gaps. Tool names lack consistent verb-noun structure (e.g., 'openclaw_message_send' vs 'send_message'). Most descriptions are adequate (50-150 chars) but lack actionable context for LLM selection. Parameter descriptions present but sparse, many lack format constraints, ranges, or dependency documentation. No output schemas documented. Error handling absent. Security concerns: no mention of credential injection, rate limiting, or permission gates. Composition issues: openclaw_sessions_send and openclaw_message_send overlap; openclaw_cron_add lacks nested schema detail.

• Document output schemas for all tools. Example: openclaw_sessions_list should return {sessions: [{sessionKey, label, kind, lastActive, messageCount}], total, nextCursor}. This enables chaining and field extraction.
• Standardize naming to verb_noun (send_message, list_sessions, spawn_agent, run_command). Rename openclaw_* prefix to match MCP convention (e.g., 'send_message' not 'openclaw_message_send').
• Add parameter constraints: openclaw_web_search.freshness as enum ['pd', 'pw', 'pm', 'py']; openclaw_nodes_camera_snap.quality as number with minimum: 0, maximum: 100; openclaw_sessions_list.limit with minimum: 1, maximum: 100.
• Add error recovery hints to descriptions. E.g., openclaw_nodes_run: 'If node not found, call openclaw_nodes_status() first to list available nodes.' openclaw_cron_run: 'Returns error if job disabled, call openclaw_cron_list to verify status.'
• Implement permission gates and audit logging. Add tool descriptions noting required scopes (e.g., 'Requires write:messages scope'). Log caller, parameters, timestamp, and result for compliance.