Critical gaps across all three tools. No input schemas are visible in the source code, all three tools lack parameter definitions, type information, and constraints. Descriptions exist but are vague and lack actionable guidance for LLM selection. No output schemas documented. Tools combine multiple responsibilities (attach_session bundles session creation + onboarding retrieval; send_message posts + journals). No error handling guidance, no recovery paths, no parameter validation rules stated.

• Add explicit JSON Schema input definitions for all three tools. At minimum: attach_session (no params or optional project/agent filters), get_session_id (no params), send_message (required: recipient_agent, message_body; optional: in_response_to, priority). Include type, description, and constraints for each parameter.
• Rewrite descriptions as LLM-optimized prompts: 'Call this FIRST to get your session ID and onboarding guidelines (memory rules, anti-sycophancy settings). Returns {session_id, project, agent, guidelines, locks, signals}. Use before any memory queries.' Current descriptions assume prior context.
• Document output schemas for all tools. Example: attach_session returns {status: 'ready'|'not_ready', session_id: string, project: string, agent: string, guidelines: [...], locks: [...], signals: [...]}. This enables downstream tool chaining.
• Split send_message into two tools: send_message (posts to inbox, returns message_id) and optionally journal_message_locally (records in local cache when server offline). Clarify which is the primary path and when journaling is automatic vs manual.
• Add error handling guidance: 'If session_id is null, the server is offline, retry attach_session in 5 seconds. If message_id is null, journaling failed, check JUNTO_API_KEY_FILE permissions.' Include recovery paths for each failure mode.