Critical failures across naming, descriptions, and schemas. Most tools lack proper verb-noun naming (e.g., 'agent-router' instead of 'route_agent', 'hive-append' instead of 'append_to_hive'). Parameter descriptions are sparse or missing entirely. Input schemas visible for only ~6 tools; most lack documented output schemas. No error handling guidance. Tools in archive/ folder suggest abandoned/orphaned code. Descriptions range 20-200 chars but many are vague (e.g., 'Claude wrapper' for tool #10). No security annotations, no permission gates, no audit trail patterns.

• Rename all tools to verb_noun pattern: agent-router → route_agent, hive-append → append_to_hive, telegram-poll → poll_telegram, etc. This enables LLMs to infer intent from names alone.
• Add explicit output schemas to all tools. Document return types, field names, and structure. E.g., route_agent returns {agent_id, name, model, effort, strategy, fallback_models}.
• Expand descriptions for wrapper tools (claude, claude-free, claude-local) to 100-150 chars explaining: (1) what model/backend it uses, (2) when to prefer it over alternatives, (3) any side effects (e.g., 'injects pending reports').
• Add error handling guidance: document what errors each tool can return, whether they are retryable, and what the LLM should do. E.g., 'If agent-router returns NO_MATCH, retry with a more specific query or escalate to A1 Lead Orchestrator.'
• Move archived tools out of tool registry or mark them as deprecated. If telegram-* tools are no longer maintained, remove them from the active tool list or add a deprecation notice to descriptions.