KAIJU provides 19 tools with explicit schemas and descriptions, but quality is inconsistent. Tool names follow verb_noun convention (dns_lookup, port_scan, file_read), which is good. However, descriptions are often terse (10-50 chars) and lack context on WHEN to use each tool or what happens on failure. Parameters have types and defaults, but many lack detailed constraints (e.g., port_scan accepts '1-1024' or 'top100' but the description doesn't explain the format). Critical security issue: command_exec and file_write are DESTRUCTIVE/WRITE tools with minimal safeguards documented in the schema itself. Output schemas are not documented, LLMs cannot predict what fields to expect from tool responses. Error handling is absent from tool definitions.

• Document output schemas for all tools. For dns_lookup, specify: {records: [{type: string, value: string, ttl: int}], query_time_ms: int}. For port_scan: {open_ports: [{port: int, service: string, banner: string}], scan_time_ms: int}. This enables LLM chaining.
• Add confirmation/dry-run support to destructive tools. command_exec should accept a 'dry_run' parameter that returns the command without executing. file_write should accept 'mode' enum (write|append|overwrite) and return a preview before committing.
• Expand descriptions to 100-200 chars with WHEN guidance. E.g., zone_transfer: 'Attempt DNS zone transfer (AXFR) to enumerate all DNS records at once. Use when you have nameserver access and want a complete zone snapshot; slower than dns_lookup but more comprehensive.'
• Formalize port_scan 'ports' parameter as enum or pattern. Document: 'Ports to scan: single (80), list (22,80,443), range (1-1024), or preset (top100|top1000). Ranges expand to all ports in range.'
• Add error recovery guidance to tool descriptions. E.g., command_exec: 'Returns exit code and stderr. Exit code 0 = success. Non-zero = command failed (retryable if timeout, user-fixable if syntax error). Timeout errors suggest increasing timeout parameter.'