HexClaw exhibits severe definition quality gaps across all 22 tools. Tool definitions are inferred from test files rather than explicitly registered in source code. Descriptions are present but generic (avg ~40 chars, well below 194-char baseline). Input schemas lack type information for most parameters, many show only string/enum without validation constraints. No output schemas documented. Critical security tools (shell, code_exec, file_ops) lack permission gates or dry-run patterns. Composition issues: file_edit and file_ops overlap; spawn_agent and orchestrate duplicate multi-agent concerns. No error recovery guidance. This server would not pass production code review.

• Explicitly register all 22 tools in main codebase with full JSON Schema definitions (type, description, enum, minLength, maxLength, pattern for each parameter). Move definitions out of test files.
• Expand tool descriptions to 100 - 200 chars following pattern:tool-description. Include WHAT, WHEN, and state-modification warnings. E.g., 'Execute shell commands on the host system. WARNING: Irreversible. Verify command before calling. Returns stdout/stderr and exit code.'
• Add output schemas documenting return types, field names, and pagination. E.g., search should return {results: [{title, url, snippet}], total_count, next_cursor}.
• Implement permission gates and audit logging for all destructive tools (shell, code_exec, file_ops, cron_task, manage_*). Add dry-run or confirmation_required patterns per pattern:confirmation-request.
• Consolidate overlapping tools: merge file_edit into file_ops with operation enum; merge spawn_agent into orchestrate with clear role separation. Document tool composition chains.