Server has 7 tools with complete input schemas and descriptions. Naming follows verb_noun pattern (it_list, it_new, it_read, it_send, it_kill, it_tail, it_head). Descriptions are present (avg ~80 chars) but lack actionable guidance on when to use each tool vs. alternatives. Parameters are typed and described, but lack constraints (enums, ranges, patterns). Output schemas are not documented in the source. Error handling returns structured errors with hints, which is good, but no recovery guidance for common failure modes. Missing: parameter constraints (e.g., lines 1-5000 for it_tail/it_head should be enforced with min/max), output schema documentation, and tool composition guidance (e.g., when to use it_read vs it_tail).

• Document output schemas for all tools. Specify what fields it_list returns (e.g., sessions: [{id, name, status}], total_count), what it_new returns (session_id, screen_content), etc. This enables LLMs to chain tools and extract needed data.
• Add min/max constraints to numeric parameters. it_tail/it_head: lines min=1, max=5000. it_new: cols min=20, max=200, rows min=5, max=100. it_list: page min=1. Encode in JSON Schema and repeat in descriptions.
• Enhance tool descriptions with selection guidance. E.g., it_read: 'Get the current terminal state (screen + scrollback). Use it_tail to read only recent lines, or it_head for initial output.' This helps LLMs pick the right tool.
• Add enum constraints for signal parameter in it_kill. signal: enum=['TERM', 'KILL', 'INT', 'HUP']. Prevents LLMs from passing invalid signals.
• Implement confirmation pattern for it_kill. Return a result with 'confirmation_required' flag and require a second call with 'confirm=true' to actually terminate. Prevents accidental session kills.